LLVM JIT interface #2277

pyos · 2018-04-24T18:40:37Z

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

This pull request adds the capability to compile built-in functions with numeric (possibly nullable) arguments and return type to native code through llvm::IRBuilder. Compilable functions are automatically inlined into each other for a performance boost. The function can then also decide where to evaluate each of its inlined arguments, allowing for some laziness (non-compilable subexpressions still have to be evaluated eagerly).

This PR also includes implementations of the interface for most arithmetic and logic functions. Don't have any performance comparisons yet, though.

Known problems:

~~this code will break with LLVM 7 when it's released due to API incompatibilities;~~
LLVM has a ton of unused parameters in its header files, so I had to add -Wno-unused-parameter to clickhouse_functions;
always inlining compilable subexpressions completely undoes all of common subexpression eliminator's work — using a heuristic, e.g. thresholding based on the size of the subexpression's graph multiplied by the number of times it is used, might be better overall;
and has somewhat weird semantics in that and(false, null) is null. This means it's impossible for and to be lazy in the second argument if it's nullable. (and(x, non-nullable) and and(null, x) work fine, though.)

Not actually implemented, though. It does print out some jit-compiled stuff, but that's about it. For example, this query: select number from system.numbers where something(cast(number as Float64)) == 4 results in this on server's stderr: define double @"something(CAST(number, 'Float64'))"(void**, i8*, void*) { "something(CAST(number, 'Float64'))": ret double 1.234500e+04 } (and an exception, because that's what the non-jitted method does.) As one may notice, this function neither reads the input (first argument; tuple of arrays) nor writes the output (third argument; array), instead returning some general nonsense. In addition, `#if USE_EMBEDDED_COMPILER` doesn't work for some reason, including LLVM headers requires -Wno-unused-parameter, this probably only works on LLVM 5.0 due to rampant API instability, and I'm definitely no expert on CMake. In short, there's still a long way to go.

The example from the previous commit doesn't need a cast to Float64 anymore.

It actually seems to work, so long as you only have one row that is. E.g. > select something(cast(number + 6 as Float64), cast(number + 2 as Float64)) from system.numbers limit 1'; 8 with this IR: define void @"something(CAST(plus(number, 6), 'Float64'), CAST(plus(number, 2), 'Float64'))"(void**, i8*, double*) { entry: %3 = load void*, void** %0 %4 = bitcast void* %3 to double* %5 = load double, double* %4 %6 = getelementptr void*, void** %0, i32 1 %7 = load void*, void** %6 %8 = bitcast void* %7 to double* %9 = load double, double* %8 %10 = fadd double %5, %9 store double %10, double* %2 ret void }

I honestly can't tell if they work. LLVM has surprisingly bad API documentation.

Given that the list of supported types is hardcoded in LLVMContext::Data::toNativeType, this method is redundant because LLVMPreparedFunction can create a ColumnVector itself.

KochetovNicolai · 2018-04-25T13:01:23Z

dbms/src/Interpreters/ExpressionActions.cpp

+    std::vector<bool> redundant(actions.size());
+    // an empty optional is a poisoned value prohibiting the column's producer from being removed
+    // (which it could be, if it was inlined into every dependent function).
+    std::unordered_map<std::string, std::unordered_set<std::optional<size_t>>> current_dependents;


It's better use struct { bool is_used; std::unordered_set<size_t> dependents; } instead of std::unordered_set<std::optional<size_t>>

proller · 2018-04-25T13:53:08Z

dbms/src/Server/Compiler-5.0.0/CMakeLists.txt

+
+if (MAKE_STATIC_LIBRARIES)
+    # fix strange static error: undefined reference to 'std::error_category::~error_category()'
+    target_link_libraries(clickhouse-compiler-lib PUBLIC stdc++)


This fix nedded only for old distributions (ubuntu trusty, xenial)
In ubuntu artful anything ok.
and this fix adds dependency to shared system lib - it unacceptable for our fully static package.

proller · 2018-04-25T13:57:20Z

dbms/CMakeLists.txt

+    # TODO: global-disable no-unused-parameter
+    set_source_files_properties(src/Interpreters/ExpressionJIT.cpp PROPERTIES COMPILE_FLAGS "-Wno-unused-parameter -Wno-non-virtual-dtor")
+else ()
+    list (REMOVE_ITEM dbms_sources src/Interpreters/ExpressionJIT.cpp)


use in this files:

#include <Common/config.h> #if USE_EMBEDDED_COMPILER ... #endif

KochetovNicolai · 2018-04-25T13:07:47Z

dbms/src/Interpreters/ExpressionJIT.cpp

+template <typename T>
+static bool typeIsA(const DataTypePtr & type)
+{
+    if (auto * nullable = typeid_cast<const DataTypeNullable *>(type.get()))


You can use removeNullable(type)

KochetovNicolai · 2018-04-25T13:10:56Z

dbms/src/Interpreters/ExpressionJIT.cpp

+
+static MutableColumnPtr createNonNullableColumn(const DataTypePtr & type)
+{
+    if (auto * nullable = typeid_cast<const DataTypeNullable *>(type.get()))


removeNullable(type)->createColumn()

KochetovNicolai · 2018-04-25T13:11:48Z

dbms/src/Interpreters/ExpressionJIT.cpp

+
+void LLVMPreparedFunction::executeImpl(Block & block, const ColumnNumbers & arguments, size_t result)
+{
+    size_t block_size = 0;


block.rows()?

KochetovNicolai · 2018-04-25T13:32:21Z

dbms/src/Interpreters/ExpressionJIT.cpp

+            /// assume the column is a `ColumnVector<T>`. there's probably no good way to actually
+            /// check that at runtime, so let's just hope it's always true for columns containing types
+            /// for which `LLVMContext::Data::toNativeType` returns non-null.
+            columns[i] = column->getDataAt(0).data;


It's very dangerous. What you can do is:

Add IColumn::isColumnVector() and StringRef IColumn::getData() implemented for columns which store data in a single continuous memory segment.
or

Use TypeListNumbers::forEach and check typeid for each ColumnVector

IColumn::isFixedAndContiguous seems to be what I wanted here. The compiled loop could be extended to arbitrary columns for which this method returns true by passing a tuple of strides (e.g. string length for ColumnFixedString) instead of a tuple of "is constant" flags, though I'm not sure how this could affect loop auto-vectorization.

…s-llvm-jit

…into llvm-jit

…vm-jit

Mostly for intrinsics like memcpy/memset/memmove, which are inserted during optimization by LLVM itself. (With a null resolver, a compiled version of something like `Uint64 < 0` would segfault.)

That one has some edge cases which I can't be bothered to code.

pyos force-pushed the llvm-jit branch from a6d63c7 to 12645b6 Compare April 24, 2018 19:43

pyos added 15 commits April 25, 2018 13:37

Add an example function that uses LLVM to compile its own body

27d90fb

Map all number types to LLVM types.

b398ffb

The example from the previous commit doesn't need a cast to Float64 anymore.

Separate jit-compilability checks from actual compilation

407008a

Implement a loop over the columns in jit-compiled code

5f1bf11

Extend the test jit-compilable function to arbitrary numbers

8c8a8f9

Enable the default set of LLVM optimization passes

6b526f7

I honestly can't tell if they work. LLVM has surprisingly bad API documentation.

Remove IFunction::createResultColumn.

3810173

Given that the list of supported types is hardcoded in LLVMContext::Data::toNativeType, this method is redundant because LLVMPreparedFunction can create a ColumnVector itself.

Inline jit-compilable functions into other jit-compilable functions

b2077a4

Tweak the jit compilation API to be more amenable to lazy computation

df2d2e0

Add missing option to CMakeFiles.txt

2b1be27

Fix CMakeFiles syntax

3789eba

Fix some comments

4bd0906

Support nullable columns (with default behavior) in jitted functions

1bece1d

pyos force-pushed the llvm-jit branch 2 times, most recently from 8b900cc to 888d97c Compare April 25, 2018 11:23

pyos added 5 commits April 25, 2018 16:01

Do not compile the jit if USE_EMBEDDED_COMPILER is disabled

0da1102

Fix some comments' style

162a0c8

Move function compilation before insertion of REMOVE_COLUMNs

af7ecd4

Implement informational methods for LLVMFunction

5482282

Poison only columns actually used by ARRAY_JOIN and JOIN

c419d5a

pyos force-pushed the llvm-jit branch from e6d7958 to c419d5a Compare April 25, 2018 13:01

KochetovNicolai reviewed Apr 25, 2018

View reviewed changes

Remove an unnoticed debug return

6c275c2

proller reviewed Apr 25, 2018

View reviewed changes

KochetovNicolai reviewed Apr 25, 2018

View reviewed changes

Add IColumn::getRawData to fixed-contiguous columns

d59b0d7

alexey-milovidov added 2 commits May 10, 2018 00:07

Fixed include paths ClickHouse#2277

9ec6bb7

Merge branch 'pyos-llvm-jit' of github.com:yandex/ClickHouse into pyo…

b8d818c

…s-llvm-jit

alexey-milovidov added a commit that referenced this pull request May 9, 2018

Fixed include paths #2277

c8b46d4

Fixed include paths ClickHouse#2277

10eae3f

alexey-milovidov added a commit that referenced this pull request May 9, 2018

Fixed include paths #2277

5197a77

alexey-milovidov and others added 11 commits May 10, 2018 00:33

Updated submodule ClickHouse#2277

a950ef6

Updated submodule ClickHouse#2277

e6369d4

Updated submodule ClickHouse#2277

3679925

Updated submodule ClickHouse#2277

3f34ece

Fixed error ClickHouse#2277

6f69c41

Updated submodule ClickHouse#2277

2d0e005

Updated submodule ClickHouse#2277

0db2e55

Merge branch 'pyos-llvm-jit' of https://github.com/yandex/ClickHouse …

4abcef1

…into llvm-jit

Merge branch 'master' of https://github.com/yandex/ClickHouse into ll…

4776168

…vm-jit

Allow calling into libc from jitted code.

bd332b9

Mostly for intrinsics like memcpy/memset/memmove, which are inserted during optimization by LLVM itself. (With a null resolver, a compiled version of something like `Uint64 < 0` would segfault.)

Implement jit for comparisons, except for (double, int).

6d2259f

That one has some edge cases which I can't be bothered to code.

pyos force-pushed the llvm-jit branch from 47652fc to 6d2259f Compare May 10, 2018 14:03

Fix incorrect phi node edges in if

c123be1

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Changed some doc files to symlinks #2277

6ecc6d8

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Minor modifications in build instruction #2277

7df26c9

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Changed some doc files to symlinks #2277

c00b5fd

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Minor modifications in build instruction #2277

01656e8

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Miscellaneous #2277

d3b7baf

pyos added 2 commits May 10, 2018 20:31

Bridge between incompatible LLVM APIs

1718e57

Revert the part of bd332b that moved a read after std::move.

e5ebc24

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Updated submodule #2277

4971811

alexey-milovidov merged commit e5ebc24 into ClickHouse:master May 10, 2018

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Be more conservative about new feature #2277

582085b

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Updated submodule #2277

9fa5952

alexey-milovidov added a commit that referenced this pull request May 10, 2018

Avoid warnings in LLVM code #2277

294b468

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

LLVM JIT interface #2277

LLVM JIT interface #2277

pyos commented Apr 24, 2018 •

edited

Loading

KochetovNicolai Apr 25, 2018

proller Apr 25, 2018 •

edited

Loading

proller Apr 25, 2018

KochetovNicolai Apr 25, 2018

KochetovNicolai Apr 25, 2018

KochetovNicolai Apr 25, 2018

KochetovNicolai Apr 25, 2018

pyos Apr 25, 2018 •

edited

Loading

LLVM JIT interface #2277

LLVM JIT interface #2277

Conversation

pyos commented Apr 24, 2018 • edited Loading

KochetovNicolai Apr 25, 2018

Choose a reason for hiding this comment

proller Apr 25, 2018 • edited Loading

Choose a reason for hiding this comment

proller Apr 25, 2018

Choose a reason for hiding this comment

KochetovNicolai Apr 25, 2018

Choose a reason for hiding this comment

KochetovNicolai Apr 25, 2018

Choose a reason for hiding this comment

KochetovNicolai Apr 25, 2018

Choose a reason for hiding this comment

KochetovNicolai Apr 25, 2018

Choose a reason for hiding this comment

pyos Apr 25, 2018 • edited Loading

Choose a reason for hiding this comment

pyos commented Apr 24, 2018 •

edited

Loading

proller Apr 25, 2018 •

edited

Loading

pyos Apr 25, 2018 •

edited

Loading