Auto-Vectorization in LLVM
==========================
-LLVM has two vectorizers: The *Loop Vectorizer*, which operates on Loops,
-and the *Basic Block Vectorizer*, which optimizes straight-line code. These
-vectorizers focus on different optimization opportunities and use different
-techniques. The BB vectorizer merges multiple scalars that are found in the
-code into vectors while the Loop Vectorizer widens instructions in the
-original loop to operate on multiple consecutive loop iterations.
+.. contents::
+ :local:
+
+LLVM has two vectorizers: The :ref:`Loop Vectorizer <loop-vectorizer>`,
+which operates on Loops, and the :ref:`SLP Vectorizer
+<slp-vectorizer>`. These vectorizers
+focus on different optimization opportunities and use different techniques.
+The SLP vectorizer merges multiple scalars that are found in the code into
+vectors while the Loop Vectorizer widens instructions in loops
+to operate on multiple consecutive iterations.
+
+Both the Loop Vectorizer and the SLP Vectorizer are enabled by default.
+
+.. _loop-vectorizer:
The Loop Vectorizer
===================
Usage
-----
-LLVM's Loop Vectorizer is now available and will be useful for many people.
-It is not enabled by default, but can be enabled through clang using the
-command line flag:
+The Loop Vectorizer is enabled by default, but it can be disabled
+through clang using the command line flag:
+
+.. code-block:: console
+
+ $ clang ... -fno-vectorize file.c
+
+Command line flags
+^^^^^^^^^^^^^^^^^^
+
+The loop vectorizer uses a cost model to decide on the optimal vectorization factor
+and unroll factor. However, users of the vectorizer can force the vectorizer to use
+specific values. Both 'clang' and 'opt' support the flags below.
+
+Users can control the vectorization SIMD width using the command line flag "-force-vector-width".
+
+.. code-block:: console
+
+ $ clang -mllvm -force-vector-width=8 ...
+ $ opt -loop-vectorize -force-vector-width=8 ...
+
+Users can control the unroll factor using the command line flag "-force-vector-interleave"
.. code-block:: console
- $ clang -fvectorize -O3 file.c
+ $ clang -mllvm -force-vector-interleave=2 ...
+ $ opt -loop-vectorize -force-vector-interleave=2 ...
+
+Pragma loop hint directives
+^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+The ``#pragma clang loop`` directive allows loop vectorization hints to be
+specified for the subsequent for, while, do-while, or c++11 range-based for
+loop. The directive allows vectorization and interleaving to be enabled or
+disabled. Vector width as well as interleave count can also be manually
+specified. The following example explicitly enables vectorization and
+interleaving:
+
+.. code-block:: c++
+
+ #pragma clang loop vectorize(enable) interleave(enable)
+ while(...) {
+ ...
+ }
-If the ``-fvectorize`` flag is used then the loop vectorizer will be enabled
-when running with ``-O3``, ``-O2``. When ``-Os`` is used, the loop vectorizer
-will only vectorize loops that do not require a major increase in code size.
+The following example implicitly enables vectorization and interleaving by
+specifying a vector width and interleaving count:
-We plan to enable the Loop Vectorizer by default as part of the LLVM 3.3 release.
+.. code-block:: c++
+
+ #pragma clang loop vectorize_width(2) interleave_count(2)
+ for(...) {
+ ...
+ }
+
+See the Clang
+`language extensions
+<http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations>`_
+for details.
+
+Diagnostics
+-----------
+
+Many loops cannot be vectorized including loops with complicated control flow,
+unvectorizable types, and unvectorizable calls. The loop vectorizer generates
+optimization remarks which can be queried using command line options to identify
+and diagnose loops that are skipped by the loop-vectorizer.
+
+Optimization remarks are enabled using:
+
+``-Rpass=loop-vectorize`` identifies loops that were successfully vectorized.
+
+``-Rpass-missed=loop-vectorize`` identifies loops that failed vectorization and
+indicates if vectorization was specified.
+
+``-Rpass-analysis=loop-vectorize`` identifies the statements that caused
+vectorization to fail. If in addition ``-fsave-optimization-record`` is
+provided, multiple causes of vectorization failure may be listed (this behavior
+might change in the future).
+
+Consider the following loop:
+
+.. code-block:: c++
+
+ #pragma clang loop vectorize(enable)
+ for (int i = 0; i < Length; i++) {
+ switch(A[i]) {
+ case 0: A[i] = i*2; break;
+ case 1: A[i] = i; break;
+ default: A[i] = 0;
+ }
+ }
+
+The command line ``-Rpass-missed=loop-vectorized`` prints the remark:
+
+.. code-block:: console
+
+ no_switch.cpp:4:5: remark: loop not vectorized: vectorization is explicitly enabled [-Rpass-missed=loop-vectorize]
+
+And the command line ``-Rpass-analysis=loop-vectorize`` indicates that the
+switch statement cannot be vectorized.
+
+.. code-block:: console
+
+ no_switch.cpp:4:5: remark: loop not vectorized: loop contains a switch statement [-Rpass-analysis=loop-vectorize]
+ switch(A[i]) {
+ ^
+
+To ensure line and column numbers are produced include the command line options
+``-gline-tables-only`` and ``-gcolumn-info``. See the Clang `user manual
+<http://clang.llvm.org/docs/UsersManual.html#options-to-emit-optimization-reports>`_
+for details
Features
--------
.. code-block:: c++
void bar(float *A, float* B, float K, int start, int end) {
- for (int i = start; i < end; ++i)
- A[i] *= B[i] + K;
+ for (int i = start; i < end; ++i)
+ A[i] *= B[i] + K;
}
Runtime Checks of Pointers
knowing that the pointers A and B are unique. The Loop Vectorizer handles this
loop by placing code that checks, at runtime, if the arrays A and B point to
disjointed memory locations. If arrays A and B overlap, then the scalar version
-of the loop is executed.
+of the loop is executed.
.. code-block:: c++
void bar(float *A, float* B, float K, int n) {
- for (int i = 0; i < n; ++i)
- A[i] *= B[i] + K;
+ for (int i = 0; i < n; ++i)
+ A[i] *= B[i] + K;
}
Reductions
^^^^^^^^^^
-In this example the ``sum`` variable is used by consecutive iterations of
+In this example the ``sum`` variable is used by consecutive iterations of
the loop. Normally, this would prevent vectorization, but the vectorizer can
detect that 'sum' is a reduction variable. The variable 'sum' becomes a vector
of integers, and at the end of the loop the elements of the array are added
-together to create the correct result. We support a number of different
+together to create the correct result. We support a number of different
reduction operations, such as addition, multiplication, XOR, AND and OR.
.. code-block:: c++
int foo(int *A, int *B, int n) {
unsigned sum = 0;
for (int i = 0; i < n; ++i)
- sum += A[i] + 5;
+ sum += A[i] + 5;
return sum;
}
+We support floating point reduction operations when `-ffast-math` is used.
+
Inductions
^^^^^^^^^^
.. code-block:: c++
void bar(float *A, float* B, float K, int n) {
- for (int i = 0; i < n; ++i)
- A[i] = i;
+ for (int i = 0; i < n; ++i)
+ A[i] = i;
}
If Conversion
Scatter / Gather
^^^^^^^^^^^^^^^^
-The Loop Vectorizer can vectorize code that becomes scatter/gather
-memory accesses.
+The Loop Vectorizer can vectorize code that becomes a sequence of scalar instructions
+that scatter/gathers memory.
.. code-block:: c++
- int foo(int *A, int *B, int n, int k) {
- for (int i = 0; i < n; ++i)
- A[i*7] += B[i*k];
+ int foo(int * A, int * B, int n) {
+ for (intptr_t i = 0; i < n; ++i)
+ A[i] += B[i * 4];
}
+In many situations the cost model will inform LLVM that this is not beneficial
+and LLVM will only vectorize such code if forced with "-mllvm -force-vector-width=#".
+
Vectorization of Mixed Types
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.. code-block:: c++
int foo(int *A, char *B, int n, int k) {
- for (int i = 0; i < n; ++i)
+ for (int i = 0; i < n; ++i)
A[i] += 4 * B[i];
}
+Global Structures Alias Analysis
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Access to global structures can also be vectorized, with alias analysis being
+used to make sure accesses don't alias. Run-time checks can also be added on
+pointer access to structure members.
+
+Many variations are supported, but some that rely on undefined behaviour being
+ignored (as other compilers do) are still being left un-vectorized.
+
+.. code-block:: c++
+
+ struct { int A[100], K, B[100]; } Foo;
+
+ int foo() {
+ for (int i = 0; i < 100; ++i)
+ Foo.A[i] = Foo.B[i] + 100;
+ }
+
Vectorization of function calls
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+-----+-----+---------+
|fma |trunc|nearbyint|
+-----+-----+---------+
+| | | fmuladd |
++-----+-----+---------+
+
+The loop vectorizer knows about special instructions on the target and will
+vectorize a loop containing a function call that maps to the instructions. For
+example, the loop below will be vectorized on Intel x86 if the SSE4.1 roundps
+instruction is available.
+
+.. code-block:: c++
+
+ void foo(float *f) {
+ for (int i = 0; i != 1024; ++i)
+ f[i] = floorf(f[i]);
+ }
+
+Partial unrolling during vectorization
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Modern processors feature multiple execution units, and only programs that contain a
+high degree of parallelism can fully utilize the entire width of the machine.
+The Loop Vectorizer increases the instruction level parallelism (ILP) by
+performing partial-unrolling of loops.
+
+In the example below the entire array is accumulated into the variable 'sum'.
+This is inefficient because only a single execution port can be used by the processor.
+By unrolling the code the Loop Vectorizer allows two or more execution ports
+to be used simultaneously.
+
+.. code-block:: c++
+
+ int foo(int *A, int *B, int n) {
+ unsigned sum = 0;
+ for (int i = 0; i < n; ++i)
+ sum += A[i];
+ return sum;
+ }
+
+The Loop Vectorizer uses a cost model to decide when it is profitable to unroll loops.
+The decision to unroll the loop depends on the register pressure and the generated code size.
Performance
-----------
-This section shows the the execution time of Clang on a simple benchmark:
+This section shows the execution time of Clang on a simple benchmark:
`gcc-loops <http://llvm.org/viewvc/llvm-project/test-suite/trunk/SingleSource/UnitTests/Vectorizer/>`_.
-This benchmarks is a collection of loops from the GCC autovectorization
+This benchmarks is a collection of loops from the GCC autovectorization
`page <http://gcc.gnu.org/projects/tree-ssa/vectorization.html>`_ by Dorit Nuzman.
The chart below compares GCC-4.7, ICC-13, and Clang-SVN with and without loop vectorization at -O3, tuned for "corei7-avx", running on a Sandybridge iMac.
-The Y-axis shows the time in msec. Lower is better. The last column shows the geomean of all the kernels.
+The Y-axis shows the time in msec. Lower is better. The last column shows the geomean of all the kernels.
.. image:: gcc-loops.png
- :width: 100%
-The Basic Block Vectorizer
-==========================
+And Linpack-pc with the same configuration. Result is Mflops, higher is better.
-Usage
-------
+.. image:: linpack-pc.png
-The Basic Block Vectorizer is not enabled by default, but it can be enabled
-through clang using the command line flag:
+Ongoing Development Directions
+------------------------------
-.. code-block:: console
+.. toctree::
+ :hidden:
+
+ Proposals/VectorizationPlan
+
+:doc:`Proposals/VectorizationPlan`
+ Modeling the process and upgrading the infrastructure of LLVM's Loop Vectorizer.
- $ clang -fslp-vectorize file.c
+.. _slp-vectorizer:
+
+The SLP Vectorizer
+==================
Details
-------
-The goal of basic-block vectorization (a.k.a. superword-level parallelism) is
-to combine similar independent instructions within simple control-flow regions
-into vector instructions. Memory accesses, arithemetic operations, comparison
-operations and some math functions can all be vectorized using this technique
-(subject to the capabilities of the target architecture).
+The goal of SLP vectorization (a.k.a. superword-level parallelism) is
+to combine similar independent instructions
+into vector instructions. Memory accesses, arithmetic operations, comparison
+operations, PHI-nodes, can all be vectorized using this technique.
For example, the following function performs very similar operations on its
inputs (a1, b1) and (a2, b2). The basic-block vectorizer may combine these
.. code-block:: c++
- int foo(int a1, int a2, int b1, int b2) {
- int r1 = a1*(a1 + b1)/b1 + 50*b1/a1;
- int r2 = a2*(a2 + b2)/b2 + 50*b2/a2;
- return r1 + r2;
+ void foo(int a1, int a2, int b1, int b2, int *A) {
+ A[0] = a1*(a1 + b1)/b1 + 50*b1/a1;
+ A[1] = a2*(a2 + b2)/b2 + 50*b2/a2;
}
+The SLP-vectorizer processes the code bottom-up, across basic blocks, in search of scalars to combine.
+
+Usage
+------
+
+The SLP Vectorizer is enabled by default, but it can be disabled
+through clang using the command line flag:
+
+.. code-block:: console
+ $ clang -fno-slp-vectorize file.c