Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 1 | ========================== |
| 2 | Auto-Vectorization in LLVM |
| 3 | ========================== |
| 4 | |
| 5 | LLVM has two vectorizers: The *Loop Vectorizer*, which operates on Loops, |
| 6 | and the *Basic Block Vectorizer*, which optimizes straight-line code. These |
| 7 | vectorizers focus on different optimization opportunities and use different |
| 8 | techniques. The BB vectorizer merges multiple scalars that are found in the |
| 9 | code into vectors while the Loop Vectorizer widens instructions in the |
| 10 | original loop to operate on multiple consecutive loop iterations. |
| 11 | |
| 12 | The Loop Vectorizer |
| 13 | =================== |
| 14 | |
Nadav Rotem | 0328f5e | 2012-12-19 18:04:44 +0000 | [diff] [blame^] | 15 | Usage |
| 16 | ^^^^^^ |
| 17 | |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 18 | LLVM’s Loop Vectorizer is now available and will be useful for many people. |
| 19 | It is not enabled by default, but can be enabled through clang using the |
| 20 | command line flag: |
| 21 | |
| 22 | .. code-block:: console |
| 23 | |
Nadav Rotem | 8f4a6cc | 2012-12-19 18:02:36 +0000 | [diff] [blame] | 24 | $ clang -fvectorize -O3 file.c |
| 25 | |
| 26 | If the ``-fvectorize`` flag is used then the loop vectorizer will be enabled |
| 27 | when running with ``-O3``, ``-O2``. When ``-Os`` is used, the loop vectorizer |
| 28 | will only vectorize loops that do not require a major increase in code size. |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 29 | |
| 30 | We plan to enable the Loop Vectorizer by default as part of the LLVM 3.3 release. |
| 31 | |
| 32 | Features |
| 33 | ^^^^^^^^^ |
| 34 | |
| 35 | The LLVM Loop Vectorizer has a number of features that allow it to vectorize |
| 36 | complex loops. |
| 37 | |
| 38 | Loops with unknown trip count |
| 39 | ------------------------------ |
| 40 | |
| 41 | The Loop Vectorizer supports loops with an unknown trip count. |
| 42 | In the loop below, the iteration ``start`` and ``finish`` points are unknown, |
| 43 | and the Loop Vectorizer has a mechanism to vectorize loops that do not start |
| 44 | at zero. In this example, ‘n’ may not be a multiple of the vector width, and |
| 45 | the vectorizer has to execute the last few iterations as scalar code. Keeping |
| 46 | a scalar copy of the loop increases the code size. |
| 47 | |
| 48 | .. code-block:: c++ |
| 49 | |
| 50 | void bar(float *A, float* B, float K, int start, int end) { |
| 51 | for (int i = start; i < end; ++i) |
| 52 | A[i] *= B[i] + K; |
| 53 | } |
| 54 | |
| 55 | Runtime Checks of Pointers |
| 56 | -------------------------- |
| 57 | |
| 58 | In the example below, if the pointers A and B point to consecutive addresses, |
| 59 | then it is illegal to vectorize the code because some elements of A will be |
| 60 | written before they are read from array B. |
| 61 | |
| 62 | Some programmers use the 'restrict' keyword to notify the compiler that the |
| 63 | pointers are disjointed, but in our example, the Loop Vectorizer has no way of |
| 64 | knowing that the pointers A and B are unique. The Loop Vectorizer handles this |
| 65 | loop by placing code that checks, at runtime, if the arrays A and B point to |
| 66 | disjointed memory locations. If arrays A and B overlap, then the scalar version |
| 67 | of the loop is executed. |
| 68 | |
| 69 | .. code-block:: c++ |
| 70 | |
| 71 | void bar(float *A, float* B, float K, int n) { |
| 72 | for (int i = 0; i < n; ++i) |
| 73 | A[i] *= B[i] + K; |
| 74 | } |
| 75 | |
| 76 | |
| 77 | Reductions |
| 78 | -------------------------- |
| 79 | |
| 80 | In this example the ``sum`` variable is used by consecutive iterations of |
| 81 | the loop. Normally, this would prevent vectorization, but the vectorizer can |
| 82 | detect that ‘sum’ is a reduction variable. The variable ‘sum’ becomes a vector |
| 83 | of integers, and at the end of the loop the elements of the array are added |
| 84 | together to create the correct result. We support a number of different |
| 85 | reduction operations, such as addition, multiplication, XOR, AND and OR. |
| 86 | |
| 87 | .. code-block:: c++ |
| 88 | |
| 89 | int foo(int *A, int *B, int n) { |
| 90 | unsigned sum = 0; |
| 91 | for (int i = 0; i < n; ++i) |
| 92 | sum += A[i] + 5; |
| 93 | return sum; |
| 94 | } |
| 95 | |
| 96 | Inductions |
| 97 | -------------------------- |
| 98 | |
| 99 | In this example the value of the induction variable ``i`` is saved into an |
| 100 | array. The Loop Vectorizer knows to vectorize induction variables. |
| 101 | |
| 102 | .. code-block:: c++ |
| 103 | |
| 104 | void bar(float *A, float* B, float K, int n) { |
| 105 | for (int i = 0; i < n; ++i) |
| 106 | A[i] = i; |
| 107 | } |
| 108 | |
| 109 | If Conversion |
| 110 | -------------------------- |
| 111 | |
| 112 | The Loop Vectorizer is able to "flatten" the IF statement in the code and |
| 113 | generate a single stream of instructions. The Loop Vectorizer supports any |
| 114 | control flow in the innermost loop. The innermost loop may contain complex |
| 115 | nesting of IFs, ELSEs and even GOTOs. |
| 116 | |
| 117 | .. code-block:: c++ |
| 118 | |
| 119 | int foo(int *A, int *B, int n) { |
| 120 | unsigned sum = 0; |
| 121 | for (int i = 0; i < n; ++i) |
| 122 | if (A[i] > B[i]) |
| 123 | sum += A[i] + 5; |
| 124 | return sum; |
| 125 | } |
| 126 | |
| 127 | Pointer Induction Variables |
Nadav Rotem | af14a3f | 2012-12-19 07:36:35 +0000 | [diff] [blame] | 128 | --------------------------- |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 129 | |
| 130 | This example uses the "accumulate" function of the standard c++ library. This |
| 131 | loop uses C++ iterators, which are pointers, and not integer indices. |
| 132 | The Loop Vectorizer detects pointer induction variables and can vectorize |
| 133 | this loop. This feature is important because many C++ programs use iterators. |
| 134 | |
| 135 | .. code-block:: c++ |
| 136 | |
| 137 | int baz(int *A, int n) { |
| 138 | return std::accumulate(A, A + n, 0); |
| 139 | } |
| 140 | |
| 141 | Reverse Iterators |
| 142 | -------------------------- |
| 143 | |
| 144 | The Loop Vectorizer can vectorize loops that count backwards. |
| 145 | |
| 146 | .. code-block:: c++ |
| 147 | |
| 148 | int foo(int *A, int *B, int n) { |
| 149 | for (int i = n; i > 0; --i) |
| 150 | A[i] +=1; |
| 151 | } |
| 152 | |
| 153 | Scatter / Gather |
Nadav Rotem | af14a3f | 2012-12-19 07:36:35 +0000 | [diff] [blame] | 154 | ---------------- |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 155 | |
Nadav Rotem | af14a3f | 2012-12-19 07:36:35 +0000 | [diff] [blame] | 156 | The Loop Vectorizer can vectorize code that becomes scatter/gather |
| 157 | memory accesses. |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 158 | |
| 159 | .. code-block:: c++ |
| 160 | |
| 161 | int foo(int *A, int *B, int n, int k) { |
| 162 | for (int i = 0; i < n; ++i) |
| 163 | A[i*7] += B[i*k]; |
| 164 | } |
| 165 | |
Nadav Rotem | af14a3f | 2012-12-19 07:36:35 +0000 | [diff] [blame] | 166 | Vectorization of Mixed Types |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 167 | -------------------------- |
| 168 | |
| 169 | The Loop Vectorizer can vectorize programs with mixed types. The Vectorizer |
| 170 | cost model can estimate the cost of the type conversion and decide if |
| 171 | vectorization is profitable. |
| 172 | |
| 173 | .. code-block:: c++ |
| 174 | |
| 175 | int foo(int *A, char *B, int n, int k) { |
| 176 | for (int i = 0; i < n; ++i) |
| 177 | A[i] += 4 * B[i]; |
| 178 | } |
| 179 | |
| 180 | Vectorization of function calls |
| 181 | -------------------------- |
| 182 | |
| 183 | The Loop Vectorize can vectorize intrinsic math functions. |
| 184 | See the table below for a list of these functions. |
| 185 | |
| 186 | +-----+-----+---------+ |
| 187 | | pow | exp | exp2 | |
| 188 | +-----+-----+---------+ |
| 189 | | sin | cos | sqrt | |
| 190 | +-----+-----+---------+ |
| 191 | | log |log2 | log10 | |
| 192 | +-----+-----+---------+ |
| 193 | |fabs |floor| ceil | |
| 194 | +-----+-----+---------+ |
| 195 | |fma |trunc|nearbyint| |
| 196 | +-----+-----+---------+ |
| 197 | |
Nadav Rotem | 15bdbbe | 2012-12-19 08:28:24 +0000 | [diff] [blame] | 198 | Performance |
| 199 | ^^^^^^^^^^^ |
| 200 | |
| 201 | This section shows the the execution time of Clang on a simple benchmark: |
Nadav Rotem | 90c8b4b | 2012-12-19 08:43:05 +0000 | [diff] [blame] | 202 | `gcc-loops <http://llvm.org/viewvc/llvm-project/test-suite/trunk/SingleSource/UnitTests/Vectorizer/>`_. |
Nadav Rotem | 15bdbbe | 2012-12-19 08:28:24 +0000 | [diff] [blame] | 203 | This benchmarks is a collection of loops from the GCC autovectorization |
Nadav Rotem | 8f4a6cc | 2012-12-19 18:02:36 +0000 | [diff] [blame] | 204 | `page <http://gcc.gnu.org/projects/tree-ssa/vectorization.html>`_ by Dorit Nuzman. |
Nadav Rotem | 15bdbbe | 2012-12-19 08:28:24 +0000 | [diff] [blame] | 205 | |
| 206 | The chart below compares GCC-4.7, ICC-13, and Clang-SVN at -O3, running on a Sandybridge. |
| 207 | The Y-axis shows time in msec. Lower is better. |
| 208 | |
| 209 | .. image:: gcc-loops.png |
| 210 | |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 211 | The Basic Block Vectorizer |
| 212 | ========================== |
| 213 | |
Nadav Rotem | 0328f5e | 2012-12-19 18:04:44 +0000 | [diff] [blame^] | 214 | Usage |
| 215 | ^^^^^^ |
| 216 | |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 217 | The Basic Block Vectorizer is not enabled by default, but it can be enabled |
| 218 | through clang using the command line flag: |
| 219 | |
| 220 | .. code-block:: console |
| 221 | |
| 222 | $ clang -fslp-vectorize file.c |
| 223 | |
Nadav Rotem | 0328f5e | 2012-12-19 18:04:44 +0000 | [diff] [blame^] | 224 | Details |
| 225 | ^^^^^^^ |
| 226 | |
Nadav Rotem | c4efbb8 | 2012-12-19 07:22:24 +0000 | [diff] [blame] | 227 | The goal of basic-block vectorization (a.k.a. superword-level parallelism) is |
| 228 | to combine similar independent instructions within simple control-flow regions |
| 229 | into vector instructions. Memory accesses, arithemetic operations, comparison |
| 230 | operations and some math functions can all be vectorized using this technique |
| 231 | (subject to the capabilities of the target architecture). |
| 232 | |
| 233 | For example, the following function performs very similar operations on its |
| 234 | inputs (a1, b1) and (a2, b2). The basic-block vectorizer may combine these |
| 235 | into vector operations. |
| 236 | |
| 237 | .. code-block:: c++ |
| 238 | |
| 239 | int foo(int a1, int a2, int b1, int b2) { |
| 240 | int r1 = a1*(a1 + b1)/b1 + 50*b1/a1; |
| 241 | int r2 = a2*(a2 + b2)/b2 + 50*b2/a2; |
| 242 | return r1 + r2; |
| 243 | } |
| 244 | |
| 245 | |