| 1 |
.. Copyright (C) 2014-2023 Free Software Foundation, Inc. |
| 2 |
Originally contributed by David Malcolm <[email protected]> |
| 3 |
|
| 4 |
This is free software: you can redistribute it and/or modify it |
| 5 |
under the terms of the GNU General Public License as published by |
| 6 |
the Free Software Foundation, either version 3 of the License, or |
| 7 |
(at your option) any later version. |
| 8 |
|
| 9 |
This program is distributed in the hope that it will be useful, but |
| 10 |
WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 12 |
General Public License for more details. |
| 13 |
|
| 14 |
You should have received a copy of the GNU General Public License |
| 15 |
along with this program. If not, see |
| 16 |
<https://www.gnu.org/licenses/>. |
| 17 |
|
| 18 |
.. default-domain:: cpp |
| 19 |
|
| 20 |
Tutorial part 3: Loops and variables |
| 21 |
------------------------------------ |
| 22 |
Consider this C function: |
| 23 |
|
| 24 |
.. code-block:: c |
| 25 |
|
| 26 |
int loop_test (int n) |
| 27 |
{ |
| 28 |
int sum = 0; |
| 29 |
for (int i = 0; i < n; i++) |
| 30 |
sum += i * i; |
| 31 |
return sum; |
| 32 |
} |
| 33 |
|
| 34 |
This example demonstrates some more features of libgccjit, with local |
| 35 |
variables and a loop. |
| 36 |
|
| 37 |
To break this down into libgccjit terms, it's usually easier to reword |
| 38 |
the `for` loop as a `while` loop, giving: |
| 39 |
|
| 40 |
.. code-block:: c |
| 41 |
|
| 42 |
int loop_test (int n) |
| 43 |
{ |
| 44 |
int sum = 0; |
| 45 |
int i = 0; |
| 46 |
while (i < n) |
| 47 |
{ |
| 48 |
sum += i * i; |
| 49 |
i++; |
| 50 |
} |
| 51 |
return sum; |
| 52 |
} |
| 53 |
|
| 54 |
Here's what the final control flow graph will look like: |
| 55 |
|
| 56 |
.. figure:: ../../intro/sum-of-squares.png |
| 57 |
:alt: image of a control flow graph |
| 58 |
|
| 59 |
As before, we include the libgccjit++ header and make a |
| 60 |
:type:`gccjit::context`. |
| 61 |
|
| 62 |
.. code-block:: c++ |
| 63 |
|
| 64 |
#include <libgccjit++.h> |
| 65 |
|
| 66 |
void test (void) |
| 67 |
{ |
| 68 |
gccjit::context ctxt; |
| 69 |
ctxt = gccjit::context::acquire (); |
| 70 |
|
| 71 |
The function works with the C `int` type. |
| 72 |
|
| 73 |
In the previous tutorial we acquired this via |
| 74 |
|
| 75 |
.. code-block:: c++ |
| 76 |
|
| 77 |
gccjit::type the_type = ctxt.get_type (ctxt, GCC_JIT_TYPE_INT); |
| 78 |
|
| 79 |
though we could equally well make it work on, say, `double`: |
| 80 |
|
| 81 |
.. code-block:: c++ |
| 82 |
|
| 83 |
gccjit::type the_type = ctxt.get_type (ctxt, GCC_JIT_TYPE_DOUBLE); |
| 84 |
|
| 85 |
For integer types we can use :func:`gccjit::context::get_int_type<T>` |
| 86 |
to directly bind a specific type: |
| 87 |
|
| 88 |
.. code-block:: c++ |
| 89 |
|
| 90 |
gccjit::type the_type = ctxt.get_int_type <int> (); |
| 91 |
|
| 92 |
Let's build the function: |
| 93 |
|
| 94 |
.. code-block:: c++ |
| 95 |
|
| 96 |
gcc_jit_param n = ctxt.new_param (the_type, "n"); |
| 97 |
std::vector<gccjit::param> params; |
| 98 |
params.push_back (n); |
| 99 |
gccjit::function func = |
| 100 |
ctxt.new_function (GCC_JIT_FUNCTION_EXPORTED, |
| 101 |
return_type, |
| 102 |
"loop_test", |
| 103 |
params, 0); |
| 104 |
|
| 105 |
Expressions: lvalues and rvalues |
| 106 |
******************************** |
| 107 |
|
| 108 |
The base class of expression is the :type:`gccjit::rvalue`, |
| 109 |
representing an expression that can be on the *right*-hand side of |
| 110 |
an assignment: a value that can be computed somehow, and assigned |
| 111 |
*to* a storage area (such as a variable). It has a specific |
| 112 |
:type:`gccjit::type`. |
| 113 |
|
| 114 |
Anothe important class is :type:`gccjit::lvalue`. |
| 115 |
A :type:`gccjit::lvalue`. is something that can of the *left*-hand |
| 116 |
side of an assignment: a storage area (such as a variable). |
| 117 |
|
| 118 |
In other words, every assignment can be thought of as: |
| 119 |
|
| 120 |
.. code-block:: c |
| 121 |
|
| 122 |
LVALUE = RVALUE; |
| 123 |
|
| 124 |
Note that :type:`gccjit::lvalue` is a subclass of |
| 125 |
:type:`gccjit::rvalue`, where in an assignment of the form: |
| 126 |
|
| 127 |
.. code-block:: c |
| 128 |
|
| 129 |
LVALUE_A = LVALUE_B; |
| 130 |
|
| 131 |
the `LVALUE_B` implies reading the current value of that storage |
| 132 |
area, assigning it into the `LVALUE_A`. |
| 133 |
|
| 134 |
So far the only expressions we've seen are from the previous tutorial: |
| 135 |
|
| 136 |
1. the multiplication `i * i`: |
| 137 |
|
| 138 |
.. code-block:: c++ |
| 139 |
|
| 140 |
gccjit::rvalue expr = |
| 141 |
ctxt.new_binary_op ( |
| 142 |
GCC_JIT_BINARY_OP_MULT, int_type, |
| 143 |
param_i, param_i); |
| 144 |
|
| 145 |
/* Alternatively, using operator-overloading: */ |
| 146 |
gccjit::rvalue expr = param_i * param_i; |
| 147 |
|
| 148 |
which is a :type:`gccjit::rvalue`, and |
| 149 |
|
| 150 |
2. the various function parameters: `param_i` and `param_n`, instances of |
| 151 |
:type:`gccjit::param`, which is a subclass of :type:`gccjit::lvalue` |
| 152 |
(and, in turn, of :type:`gccjit::rvalue`): |
| 153 |
we can both read from and write to function parameters within the |
| 154 |
body of a function. |
| 155 |
|
| 156 |
Our new example has a new kind of expression: we have two local |
| 157 |
variables. We create them by calling |
| 158 |
:func:`gccjit::function::new_local`, supplying a type and a name: |
| 159 |
|
| 160 |
.. code-block:: c++ |
| 161 |
|
| 162 |
/* Build locals: */ |
| 163 |
gccjit::lvalue i = func.new_local (the_type, "i"); |
| 164 |
gccjit::lvalue sum = func.new_local (the_type, "sum"); |
| 165 |
|
| 166 |
These are instances of :type:`gccjit::lvalue` - they can be read from |
| 167 |
and written to. |
| 168 |
|
| 169 |
Note that there is no precanned way to create *and* initialize a variable |
| 170 |
like in C: |
| 171 |
|
| 172 |
.. code-block:: c |
| 173 |
|
| 174 |
int i = 0; |
| 175 |
|
| 176 |
Instead, having added the local to the function, we have to separately add |
| 177 |
an assignment of `0` to `local_i` at the beginning of the function. |
| 178 |
|
| 179 |
Control flow |
| 180 |
************ |
| 181 |
|
| 182 |
This function has a loop, so we need to build some basic blocks to |
| 183 |
handle the control flow. In this case, we need 4 blocks: |
| 184 |
|
| 185 |
1. before the loop (initializing the locals) |
| 186 |
2. the conditional at the top of the loop (comparing `i < n`) |
| 187 |
3. the body of the loop |
| 188 |
4. after the loop terminates (`return sum`) |
| 189 |
|
| 190 |
so we create these as :type:`gccjit::block` instances within the |
| 191 |
:type:`gccjit::function`: |
| 192 |
|
| 193 |
.. code-block:: c++ |
| 194 |
|
| 195 |
gccjit::block b_initial = func.new_block ("initial"); |
| 196 |
gccjit::block b_loop_cond = func.new_block ("loop_cond"); |
| 197 |
gccjit::block b_loop_body = func.new_block ("loop_body"); |
| 198 |
gccjit::block b_after_loop = func.new_block ("after_loop"); |
| 199 |
|
| 200 |
We now populate each block with statements. |
| 201 |
|
| 202 |
The entry block `b_initial` consists of initializations followed by a jump |
| 203 |
to the conditional. We assign `0` to `i` and to `sum`, using |
| 204 |
:func:`gccjit::block::add_assignment` to add |
| 205 |
an assignment statement, and using :func:`gccjit::context::zero` to get |
| 206 |
the constant value `0` for the relevant type for the right-hand side of |
| 207 |
the assignment: |
| 208 |
|
| 209 |
.. code-block:: c++ |
| 210 |
|
| 211 |
/* sum = 0; */ |
| 212 |
b_initial.add_assignment (sum, ctxt.zero (the_type)); |
| 213 |
|
| 214 |
/* i = 0; */ |
| 215 |
b_initial.add_assignment (i, ctxt.zero (the_type)); |
| 216 |
|
| 217 |
We can then terminate the entry block by jumping to the conditional: |
| 218 |
|
| 219 |
.. code-block:: c++ |
| 220 |
|
| 221 |
b_initial.end_with_jump (b_loop_cond); |
| 222 |
|
| 223 |
The conditional block is equivalent to the line `while (i < n)` from our |
| 224 |
C example. It contains a single statement: a conditional, which jumps to |
| 225 |
one of two destination blocks depending on a boolean |
| 226 |
:type:`gccjit::rvalue`, in this case the comparison of `i` and `n`. |
| 227 |
|
| 228 |
We could build the comparison using :func:`gccjit::context::new_comparison`: |
| 229 |
|
| 230 |
.. code-block:: c++ |
| 231 |
|
| 232 |
gccjit::rvalue guard = |
| 233 |
ctxt.new_comparison (GCC_JIT_COMPARISON_GE, |
| 234 |
i, n); |
| 235 |
|
| 236 |
and can then use this to add `b_loop_cond`'s sole statement, via |
| 237 |
:func:`gccjit::block::end_with_conditional`: |
| 238 |
|
| 239 |
.. code-block:: c++ |
| 240 |
|
| 241 |
b_loop_cond.end_with_conditional (guard, |
| 242 |
b_after_loop, // on_true |
| 243 |
b_loop_body); // on_false |
| 244 |
|
| 245 |
However :type:`gccjit::rvalue` has overloaded operators for this, so we |
| 246 |
express the conditional as |
| 247 |
|
| 248 |
.. code-block:: c++ |
| 249 |
|
| 250 |
gccjit::rvalue guard = (i >= n); |
| 251 |
|
| 252 |
and hence we can write the block more concisely as: |
| 253 |
|
| 254 |
.. code-block:: c++ |
| 255 |
|
| 256 |
b_loop_cond.end_with_conditional ( |
| 257 |
i >= n, |
| 258 |
b_after_loop, // on_true |
| 259 |
b_loop_body); // on_false |
| 260 |
|
| 261 |
Next, we populate the body of the loop. |
| 262 |
|
| 263 |
The C statement `sum += i * i;` is an assignment operation, where an |
| 264 |
lvalue is modified "in-place". We use |
| 265 |
:func:`gccjit::block::add_assignment_op` to handle these operations: |
| 266 |
|
| 267 |
.. code-block:: c++ |
| 268 |
|
| 269 |
/* sum += i * i */ |
| 270 |
b_loop_body.add_assignment_op (sum, |
| 271 |
GCC_JIT_BINARY_OP_PLUS, |
| 272 |
i * i); |
| 273 |
|
| 274 |
The `i++` can be thought of as `i += 1`, and can thus be handled in |
| 275 |
a similar way. We use :c:func:`gcc_jit_context_one` to get the constant |
| 276 |
value `1` (for the relevant type) for the right-hand side |
| 277 |
of the assignment. |
| 278 |
|
| 279 |
.. code-block:: c++ |
| 280 |
|
| 281 |
/* i++ */ |
| 282 |
b_loop_body.add_assignment_op (i, |
| 283 |
GCC_JIT_BINARY_OP_PLUS, |
| 284 |
ctxt.one (the_type)); |
| 285 |
|
| 286 |
.. note:: |
| 287 |
|
| 288 |
For numeric constants other than 0 or 1, we could use |
| 289 |
:func:`gccjit::context::new_rvalue`, which has overloads |
| 290 |
for both ``int`` and ``double``. |
| 291 |
|
| 292 |
The loop body completes by jumping back to the conditional: |
| 293 |
|
| 294 |
.. code-block:: c++ |
| 295 |
|
| 296 |
b_loop_body.end_with_jump (b_loop_cond); |
| 297 |
|
| 298 |
Finally, we populate the `b_after_loop` block, reached when the loop |
| 299 |
conditional is false. We want to generate the equivalent of: |
| 300 |
|
| 301 |
.. code-block:: c++ |
| 302 |
|
| 303 |
return sum; |
| 304 |
|
| 305 |
so the block is just one statement: |
| 306 |
|
| 307 |
.. code-block:: c++ |
| 308 |
|
| 309 |
/* return sum */ |
| 310 |
b_after_loop.end_with_return (sum); |
| 311 |
|
| 312 |
.. note:: |
| 313 |
|
| 314 |
You can intermingle block creation with statement creation, |
| 315 |
but given that the terminator statements generally include references |
| 316 |
to other blocks, I find it's clearer to create all the blocks, |
| 317 |
*then* all the statements. |
| 318 |
|
| 319 |
We've finished populating the function. As before, we can now compile it |
| 320 |
to machine code: |
| 321 |
|
| 322 |
.. code-block:: c++ |
| 323 |
|
| 324 |
gcc_jit_result *result; |
| 325 |
result = ctxt.compile (); |
| 326 |
|
| 327 |
ctxt.release (); |
| 328 |
|
| 329 |
if (!result) |
| 330 |
{ |
| 331 |
fprintf (stderr, "NULL result"); |
| 332 |
return 1; |
| 333 |
} |
| 334 |
|
| 335 |
typedef int (*loop_test_fn_type) (int); |
| 336 |
loop_test_fn_type loop_test = |
| 337 |
(loop_test_fn_type)gcc_jit_result_get_code (result, "loop_test"); |
| 338 |
if (!loop_test) |
| 339 |
{ |
| 340 |
fprintf (stderr, "NULL loop_test"); |
| 341 |
gcc_jit_result_release (result); |
| 342 |
return 1; |
| 343 |
} |
| 344 |
printf ("result: %d", loop_test (10)); |
| 345 |
|
| 346 |
.. code-block:: bash |
| 347 |
|
| 348 |
result: 285 |
| 349 |
|
| 350 |
|
| 351 |
Visualizing the control flow graph |
| 352 |
********************************** |
| 353 |
|
| 354 |
You can see the control flow graph of a function using |
| 355 |
:func:`gccjit::function::dump_to_dot`: |
| 356 |
|
| 357 |
.. code-block:: c++ |
| 358 |
|
| 359 |
func.dump_to_dot ("/tmp/sum-of-squares.dot"); |
| 360 |
|
| 361 |
giving a .dot file in GraphViz format. |
| 362 |
|
| 363 |
You can convert this to an image using `dot`: |
| 364 |
|
| 365 |
.. code-block:: bash |
| 366 |
|
| 367 |
$ dot -Tpng /tmp/sum-of-squares.dot -o /tmp/sum-of-squares.png |
| 368 |
|
| 369 |
or use a viewer (my preferred one is xdot.py; see |
| 370 |
https://github.com/jrfonseca/xdot.py; on Fedora you can |
| 371 |
install it with `yum install python-xdot`): |
| 372 |
|
| 373 |
.. figure:: ../../intro/sum-of-squares.png |
| 374 |
:alt: image of a control flow graph |
| 375 |
|
| 376 |
Full example |
| 377 |
************ |
| 378 |
|
| 379 |
.. literalinclude:: ../../examples/tut03-sum-of-squares.cc |
| 380 |
:lines: 1- |
| 381 |
:language: c++ |
| 382 |
|
| 383 |
Building and running it: |
| 384 |
|
| 385 |
.. code-block:: console |
| 386 |
|
| 387 |
$ gcc \ |
| 388 |
tut03-sum-of-squares.cc \ |
| 389 |
-o tut03-sum-of-squares \ |
| 390 |
-lgccjit |
| 391 |
|
| 392 |
# Run the built program: |
| 393 |
$ ./tut03-sum-of-squares |
| 394 |
loop_test returned: 285 |