torch-mlir/lib/RefBackend/Runtime/CompilerRuntime.cpp

125 lines
4.3 KiB
C++
Raw Normal View History

Rework e2e flow to use new "npcomprt" This ~totally reworks the existing "runtime" stuff to be more principled and usable, such as from Python. It's still not fully production-quality, mainly in the department of memory management (e.g. it currently leaks memory; we need to figure out "who frees memrefs" + the analysis and transformation needed to do that (maybe use upstream buffer allocation pass?)). The user API is in include/npcomp/runtime/UserAPI.h, though include/npcomp/JITRuntime/JITModule.h is a friendlier wrapper. The stuff under {include,lib}/runtime is totally firewalled from the compiler and tiny (<6kB, though no attention has gone into optimizing that size). For example, we don't link in libSupport into the runtime, instead having our own bare bones replacements for basics like ArrayRef (the JITRuntime helps with bridging that gap, since it *can* depend on all common LLVM utilities). The overall features of npcomprt is that it exposes a module that with multiple function entry points. Each function has arguments and results that are tensor-valued, and npcomprt::Tensor is the runtime type that is used to interact with that (and a npcomprt::Ref<T> reference-counting wrapper is provided to wrap npcomprt::Tensor in the common case). From an implementation perspective, an npcomprt module at the LLVM/object/binary level exposes a single module descriptor struct that has pointers to other metadata (currently just a list of function metadata descriptors). All interactions with the npcomp runtime are keyed off of that module descriptor, including function lookups and dispatching. This is done to dodge platform ABI issues and also allow enough reflection to e.g. verify provided arguments. Most of the compiler-side work here was in LowerToNpcomprtABI and LowerToLLVM. Also, - Rename npcomp_rt/NpcompRt to npcomprt/Npcomprt; it was getting annoying to type the underscores/caps. - misc improvements to bash_helpers.sh
2020-07-09 08:15:40 +08:00
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// Symbols referenced only by the compiler and which will be compiled into a
// shared object that a JIT can load to provide those symbols.
//
//===----------------------------------------------------------------------===//
#include <array>
#include <cstdlib>
#include <iostream>
#include "CompilerDataStructures.h"
#include "npcomp/RefBackend/Runtime/UserAPI.h"
Rework e2e flow to use new "npcomprt" This ~totally reworks the existing "runtime" stuff to be more principled and usable, such as from Python. It's still not fully production-quality, mainly in the department of memory management (e.g. it currently leaks memory; we need to figure out "who frees memrefs" + the analysis and transformation needed to do that (maybe use upstream buffer allocation pass?)). The user API is in include/npcomp/runtime/UserAPI.h, though include/npcomp/JITRuntime/JITModule.h is a friendlier wrapper. The stuff under {include,lib}/runtime is totally firewalled from the compiler and tiny (<6kB, though no attention has gone into optimizing that size). For example, we don't link in libSupport into the runtime, instead having our own bare bones replacements for basics like ArrayRef (the JITRuntime helps with bridging that gap, since it *can* depend on all common LLVM utilities). The overall features of npcomprt is that it exposes a module that with multiple function entry points. Each function has arguments and results that are tensor-valued, and npcomprt::Tensor is the runtime type that is used to interact with that (and a npcomprt::Ref<T> reference-counting wrapper is provided to wrap npcomprt::Tensor in the common case). From an implementation perspective, an npcomprt module at the LLVM/object/binary level exposes a single module descriptor struct that has pointers to other metadata (currently just a list of function metadata descriptors). All interactions with the npcomp runtime are keyed off of that module descriptor, including function lookups and dispatching. This is done to dodge platform ABI issues and also allow enough reflection to e.g. verify provided arguments. Most of the compiler-side work here was in LowerToNpcomprtABI and LowerToLLVM. Also, - Rename npcomp_rt/NpcompRt to npcomprt/Npcomprt; it was getting annoying to type the underscores/caps. - misc improvements to bash_helpers.sh
2020-07-09 08:15:40 +08:00
using namespace refbackrt;
Rework e2e flow to use new "npcomprt" This ~totally reworks the existing "runtime" stuff to be more principled and usable, such as from Python. It's still not fully production-quality, mainly in the department of memory management (e.g. it currently leaks memory; we need to figure out "who frees memrefs" + the analysis and transformation needed to do that (maybe use upstream buffer allocation pass?)). The user API is in include/npcomp/runtime/UserAPI.h, though include/npcomp/JITRuntime/JITModule.h is a friendlier wrapper. The stuff under {include,lib}/runtime is totally firewalled from the compiler and tiny (<6kB, though no attention has gone into optimizing that size). For example, we don't link in libSupport into the runtime, instead having our own bare bones replacements for basics like ArrayRef (the JITRuntime helps with bridging that gap, since it *can* depend on all common LLVM utilities). The overall features of npcomprt is that it exposes a module that with multiple function entry points. Each function has arguments and results that are tensor-valued, and npcomprt::Tensor is the runtime type that is used to interact with that (and a npcomprt::Ref<T> reference-counting wrapper is provided to wrap npcomprt::Tensor in the common case). From an implementation perspective, an npcomprt module at the LLVM/object/binary level exposes a single module descriptor struct that has pointers to other metadata (currently just a list of function metadata descriptors). All interactions with the npcomp runtime are keyed off of that module descriptor, including function lookups and dispatching. This is done to dodge platform ABI issues and also allow enough reflection to e.g. verify provided arguments. Most of the compiler-side work here was in LowerToNpcomprtABI and LowerToLLVM. Also, - Rename npcomp_rt/NpcompRt to npcomprt/Npcomprt; it was getting annoying to type the underscores/caps. - misc improvements to bash_helpers.sh
2020-07-09 08:15:40 +08:00
Totally rework RefE2E tensor to memref flow. (#42) This now gets the overall "RefE2E" compilation stack to a point that I'm fairly happy with. We simplify it by mostly embracing the "descriptor" view of the world. The overall flow is best understood by reading through the createE2ELoweringPipeline function in lib/E2E/E2E.cpp That function creates a pass pipeline that lowers from "TCF" (which is ~numpy level of abstraction) down to LLVM IR. A brief high-level summary of what happens there: 1. TCF to TCP conversion. This involves reifying error handling in the form of shape constraints. See test/Conversion/TCFToTCP/basic.mlir 2. Lowering shape constraints. This converts shape constraints into eager error-handling code. See test/E2E/lower-shape-constraints.mlir This pass will soon go upstream. Because this lowers to std.assert, some later passes like LowerToNpcomprtABI and LowerToLLVM are updated to properly plumb this through e2e. See test/npcomp-run-mlir/invalid-broadcast.mlir for an execution test that properly aborts in case of an error. 3. Lowering tensors to memrefs. This is done via a series of passes rather than an single mega conversion. Unlike the previous code that mixed in the npcomprt ABI stuff here, it's now a very clean "pure memref" conversion. See test/E2E/lower-*-to-memref.mlir and lib/E2E/TensorToMemref/ Most of the changes are concentrated here. 4. As part of the above, we use the upstream ConvertShapeToStandard for lowering shapes. 5. We lower linalg to loops and lower loops to CFG using upstream passes. 6. Rewrite the "ABI" boundaries of the program to npcomprt data structures (LowerToNpcomprtABI). This mainly affects ABI boundaries and how global tensor constants are represented. One of the major improvements in this commit is that now it's a very clean rewrite that just replaces memrefs on ABI boundaries with !npcomprt.tensor (before there was a get_extent function that is not needed). See test/E2E/lower-to-npcomprt-abi.mlir 7. Lower to LLVM with upstream mlir patterns + some patterns for the npcomprt lowerings. One aspect here that is still a remnant of a non-descriptor-based tensor to memref flow is the BypassShapes + LowerShapedResultsToMemref. BypassShapes wraps the "tensor compute" ops in a tcp.shaped_results (basically a "tie_shape" kind of op), and then LowerShapedResultsToMemref uses those annotations to allocate output buffers while lowering the "tensor compute ops". Note that there are very few "tensor compute" ops currently supported (tcp.add + tcp.broadcast_to), so we just hardcode them in both passes. Realistically, I expect this to go away as we fully embrace the descriptor-based approach for simplicity, so don't look too deep into it.
2020-09-17 08:31:40 +08:00
extern "C" void __npcomp_compiler_rt_abort_if(bool b, const char *msg) {
Rework e2e flow to use new "npcomprt" This ~totally reworks the existing "runtime" stuff to be more principled and usable, such as from Python. It's still not fully production-quality, mainly in the department of memory management (e.g. it currently leaks memory; we need to figure out "who frees memrefs" + the analysis and transformation needed to do that (maybe use upstream buffer allocation pass?)). The user API is in include/npcomp/runtime/UserAPI.h, though include/npcomp/JITRuntime/JITModule.h is a friendlier wrapper. The stuff under {include,lib}/runtime is totally firewalled from the compiler and tiny (<6kB, though no attention has gone into optimizing that size). For example, we don't link in libSupport into the runtime, instead having our own bare bones replacements for basics like ArrayRef (the JITRuntime helps with bridging that gap, since it *can* depend on all common LLVM utilities). The overall features of npcomprt is that it exposes a module that with multiple function entry points. Each function has arguments and results that are tensor-valued, and npcomprt::Tensor is the runtime type that is used to interact with that (and a npcomprt::Ref<T> reference-counting wrapper is provided to wrap npcomprt::Tensor in the common case). From an implementation perspective, an npcomprt module at the LLVM/object/binary level exposes a single module descriptor struct that has pointers to other metadata (currently just a list of function metadata descriptors). All interactions with the npcomp runtime are keyed off of that module descriptor, including function lookups and dispatching. This is done to dodge platform ABI issues and also allow enough reflection to e.g. verify provided arguments. Most of the compiler-side work here was in LowerToNpcomprtABI and LowerToLLVM. Also, - Rename npcomp_rt/NpcompRt to npcomprt/Npcomprt; it was getting annoying to type the underscores/caps. - misc improvements to bash_helpers.sh
2020-07-09 08:15:40 +08:00
if (b) {
Totally rework RefE2E tensor to memref flow. (#42) This now gets the overall "RefE2E" compilation stack to a point that I'm fairly happy with. We simplify it by mostly embracing the "descriptor" view of the world. The overall flow is best understood by reading through the createE2ELoweringPipeline function in lib/E2E/E2E.cpp That function creates a pass pipeline that lowers from "TCF" (which is ~numpy level of abstraction) down to LLVM IR. A brief high-level summary of what happens there: 1. TCF to TCP conversion. This involves reifying error handling in the form of shape constraints. See test/Conversion/TCFToTCP/basic.mlir 2. Lowering shape constraints. This converts shape constraints into eager error-handling code. See test/E2E/lower-shape-constraints.mlir This pass will soon go upstream. Because this lowers to std.assert, some later passes like LowerToNpcomprtABI and LowerToLLVM are updated to properly plumb this through e2e. See test/npcomp-run-mlir/invalid-broadcast.mlir for an execution test that properly aborts in case of an error. 3. Lowering tensors to memrefs. This is done via a series of passes rather than an single mega conversion. Unlike the previous code that mixed in the npcomprt ABI stuff here, it's now a very clean "pure memref" conversion. See test/E2E/lower-*-to-memref.mlir and lib/E2E/TensorToMemref/ Most of the changes are concentrated here. 4. As part of the above, we use the upstream ConvertShapeToStandard for lowering shapes. 5. We lower linalg to loops and lower loops to CFG using upstream passes. 6. Rewrite the "ABI" boundaries of the program to npcomprt data structures (LowerToNpcomprtABI). This mainly affects ABI boundaries and how global tensor constants are represented. One of the major improvements in this commit is that now it's a very clean rewrite that just replaces memrefs on ABI boundaries with !npcomprt.tensor (before there was a get_extent function that is not needed). See test/E2E/lower-to-npcomprt-abi.mlir 7. Lower to LLVM with upstream mlir patterns + some patterns for the npcomprt lowerings. One aspect here that is still a remnant of a non-descriptor-based tensor to memref flow is the BypassShapes + LowerShapedResultsToMemref. BypassShapes wraps the "tensor compute" ops in a tcp.shaped_results (basically a "tie_shape" kind of op), and then LowerShapedResultsToMemref uses those annotations to allocate output buffers while lowering the "tensor compute ops". Note that there are very few "tensor compute" ops currently supported (tcp.add + tcp.broadcast_to), so we just hardcode them in both passes. Realistically, I expect this to go away as we fully embrace the descriptor-based approach for simplicity, so don't look too deep into it.
2020-09-17 08:31:40 +08:00
std::fprintf(stderr, "NPCOMP: aborting: %s\n", msg);
Rework e2e flow to use new "npcomprt" This ~totally reworks the existing "runtime" stuff to be more principled and usable, such as from Python. It's still not fully production-quality, mainly in the department of memory management (e.g. it currently leaks memory; we need to figure out "who frees memrefs" + the analysis and transformation needed to do that (maybe use upstream buffer allocation pass?)). The user API is in include/npcomp/runtime/UserAPI.h, though include/npcomp/JITRuntime/JITModule.h is a friendlier wrapper. The stuff under {include,lib}/runtime is totally firewalled from the compiler and tiny (<6kB, though no attention has gone into optimizing that size). For example, we don't link in libSupport into the runtime, instead having our own bare bones replacements for basics like ArrayRef (the JITRuntime helps with bridging that gap, since it *can* depend on all common LLVM utilities). The overall features of npcomprt is that it exposes a module that with multiple function entry points. Each function has arguments and results that are tensor-valued, and npcomprt::Tensor is the runtime type that is used to interact with that (and a npcomprt::Ref<T> reference-counting wrapper is provided to wrap npcomprt::Tensor in the common case). From an implementation perspective, an npcomprt module at the LLVM/object/binary level exposes a single module descriptor struct that has pointers to other metadata (currently just a list of function metadata descriptors). All interactions with the npcomp runtime are keyed off of that module descriptor, including function lookups and dispatching. This is done to dodge platform ABI issues and also allow enough reflection to e.g. verify provided arguments. Most of the compiler-side work here was in LowerToNpcomprtABI and LowerToLLVM. Also, - Rename npcomp_rt/NpcompRt to npcomprt/Npcomprt; it was getting annoying to type the underscores/caps. - misc improvements to bash_helpers.sh
2020-07-09 08:15:40 +08:00
std::exit(1);
}
}
namespace {
// These definitions are based on the ones in
// `mlir/ExecutionEngine/CRunnerUtils.h` and the layouts need to be kept in
// sync.
//
// Those definitions are flawed though because they are overly templated.
struct MemrefDescriptor {
void *allocatedPtr;
void *dataPtr;
std::int64_t offset;
// Tail-allocated int64_t sizes followed by strides.
MutableArrayRef<std::int64_t> getSizes(int assumedRank) {
auto *tail = reinterpret_cast<std::int64_t *>(this + 1);
return MutableArrayRef<std::int64_t>(tail, assumedRank);
}
MutableArrayRef<std::int64_t> getStrides(int assumedRank) {
auto *tail = reinterpret_cast<std::int64_t *>(this + 1);
return MutableArrayRef<std::int64_t>(tail + assumedRank, assumedRank);
}
// Returns a malloc-allocated MemrefDescriptor with the specified extents and
// default striding.
static MemrefDescriptor *create(ArrayRef<std::int32_t> extents, void *data);
};
struct UnrankedMemref {
int64_t rank;
MemrefDescriptor *descriptor;
};
MemrefDescriptor *MemrefDescriptor::create(ArrayRef<std::int32_t> extents,
void *data) {
auto rank = extents.size();
auto allocSize = sizeof(MemrefDescriptor) + sizeof(std::int64_t) * 2 * rank;
auto *descriptor = static_cast<MemrefDescriptor *>(std::malloc(allocSize));
descriptor->allocatedPtr = data;
descriptor->dataPtr = data;
descriptor->offset = 0;
// Iterate in reverse, copying the dimension sizes (i.e. extents) and
// calculating the strides for a standard dense layout.
std::int64_t stride = 1;
for (int i = 0, e = rank; i < e; i++) {
auto revIdx = e - i - 1;
descriptor->getSizes(rank)[revIdx] = extents[revIdx];
descriptor->getStrides(rank)[revIdx] = stride;
stride *= extents[revIdx];
}
return descriptor;
}
std::int32_t getNumElements(MemrefDescriptor *descriptor, int assumedRank) {
if (assumedRank == 0)
return 1;
return descriptor->getSizes(assumedRank)[0] *
descriptor->getStrides(assumedRank)[0];
}
} // namespace
extern "C" UnrankedMemref __npcomp_compiler_rt_to_memref(Tensor *tensor) {
auto byteSize = tensor->getDataByteSize();
void *data = std::malloc(byteSize);
std::memcpy(data, tensor->getData(), byteSize);
auto *descriptor = MemrefDescriptor::create(tensor->getExtents(), data);
return UnrankedMemref{tensor->getRank(), descriptor};
}
extern "C" Tensor *
__npcomp_compiler_rt_from_memref(std::int64_t rank,
MemrefDescriptor *descriptor) {
auto numElements = getNumElements(descriptor, rank);
// TODO: Have the compiler pass this as an argument.
auto elementType = ElementType::F32;
auto byteSize = getElementTypeByteSize(elementType) * numElements;
void *data = std::malloc(byteSize);
std::memcpy(data, descriptor->dataPtr, byteSize);
auto extents64 = descriptor->getSizes(rank);
// Launder from std::int64_t to std::int32_t.
constexpr int kMaxRank = 20;
std::array<std::int32_t, kMaxRank> extents32Buf;
for (int i = 0, e = extents64.size(); i < e; i++)
extents32Buf[i] = extents64[i];
return Tensor::createRaw(ArrayRef<std::int32_t>(extents32Buf.data(), rank),
elementType, data);
}
extern "C" UnrankedMemref
__npcomp_compiler_rt_get_global(GlobalDescriptor *global) {
auto *descriptor = MemrefDescriptor::create(
ArrayRef<std::int32_t>(global->extents, global->numExtents),
global->data);
return UnrankedMemref{global->numExtents, descriptor};
}