1 |
12 |
jlechner |
/* Lambda matrix and vector interface.
|
2 |
|
|
Copyright (C) 2003, 2004, 2005 Free Software Foundation, Inc.
|
3 |
|
|
Contributed by Daniel Berlin <dberlin@dberlin.org>
|
4 |
|
|
|
5 |
|
|
This file is part of GCC.
|
6 |
|
|
|
7 |
|
|
GCC is free software; you can redistribute it and/or modify it under
|
8 |
|
|
the terms of the GNU General Public License as published by the Free
|
9 |
|
|
Software Foundation; either version 2, or (at your option) any later
|
10 |
|
|
version.
|
11 |
|
|
|
12 |
|
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
13 |
|
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
14 |
|
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
15 |
|
|
for more details.
|
16 |
|
|
|
17 |
|
|
You should have received a copy of the GNU General Public License
|
18 |
|
|
along with GCC; see the file COPYING. If not, write to the Free
|
19 |
|
|
Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
|
20 |
|
|
02110-1301, USA. */
|
21 |
|
|
|
22 |
|
|
#ifndef LAMBDA_H
|
23 |
|
|
#define LAMBDA_H
|
24 |
|
|
|
25 |
|
|
#include "vec.h"
|
26 |
|
|
|
27 |
|
|
/* An integer vector. A vector formally consists of an element of a vector
|
28 |
|
|
space. A vector space is a set that is closed under vector addition
|
29 |
|
|
and scalar multiplication. In this vector space, an element is a list of
|
30 |
|
|
integers. */
|
31 |
|
|
typedef int *lambda_vector;
|
32 |
|
|
|
33 |
|
|
DEF_VEC_P(lambda_vector);
|
34 |
|
|
DEF_VEC_ALLOC_P(lambda_vector,heap);
|
35 |
|
|
|
36 |
|
|
/* An integer matrix. A matrix consists of m vectors of length n (IE
|
37 |
|
|
all vectors are the same length). */
|
38 |
|
|
typedef lambda_vector *lambda_matrix;
|
39 |
|
|
|
40 |
|
|
/* A transformation matrix, which is a self-contained ROWSIZE x COLSIZE
|
41 |
|
|
matrix. Rather than use floats, we simply keep a single DENOMINATOR that
|
42 |
|
|
represents the denominator for every element in the matrix. */
|
43 |
|
|
typedef struct
|
44 |
|
|
{
|
45 |
|
|
lambda_matrix matrix;
|
46 |
|
|
int rowsize;
|
47 |
|
|
int colsize;
|
48 |
|
|
int denominator;
|
49 |
|
|
} *lambda_trans_matrix;
|
50 |
|
|
#define LTM_MATRIX(T) ((T)->matrix)
|
51 |
|
|
#define LTM_ROWSIZE(T) ((T)->rowsize)
|
52 |
|
|
#define LTM_COLSIZE(T) ((T)->colsize)
|
53 |
|
|
#define LTM_DENOMINATOR(T) ((T)->denominator)
|
54 |
|
|
|
55 |
|
|
/* A vector representing a statement in the body of a loop.
|
56 |
|
|
The COEFFICIENTS vector contains a coefficient for each induction variable
|
57 |
|
|
in the loop nest containing the statement.
|
58 |
|
|
The DENOMINATOR represents the denominator for each coefficient in the
|
59 |
|
|
COEFFICIENT vector.
|
60 |
|
|
|
61 |
|
|
This structure is used during code generation in order to rewrite the old
|
62 |
|
|
induction variable uses in a statement in terms of the newly created
|
63 |
|
|
induction variables. */
|
64 |
|
|
typedef struct
|
65 |
|
|
{
|
66 |
|
|
lambda_vector coefficients;
|
67 |
|
|
int size;
|
68 |
|
|
int denominator;
|
69 |
|
|
} *lambda_body_vector;
|
70 |
|
|
#define LBV_COEFFICIENTS(T) ((T)->coefficients)
|
71 |
|
|
#define LBV_SIZE(T) ((T)->size)
|
72 |
|
|
#define LBV_DENOMINATOR(T) ((T)->denominator)
|
73 |
|
|
|
74 |
|
|
/* Piecewise linear expression.
|
75 |
|
|
This structure represents a linear expression with terms for the invariants
|
76 |
|
|
and induction variables of a loop.
|
77 |
|
|
COEFFICIENTS is a vector of coefficients for the induction variables, one
|
78 |
|
|
per loop in the loop nest.
|
79 |
|
|
CONSTANT is the constant portion of the linear expression
|
80 |
|
|
INVARIANT_COEFFICIENTS is a vector of coefficients for the loop invariants,
|
81 |
|
|
one per invariant.
|
82 |
|
|
DENOMINATOR is the denominator for all of the coefficients and constants in
|
83 |
|
|
the expression.
|
84 |
|
|
The linear expressions can be linked together using the NEXT field, in
|
85 |
|
|
order to represent MAX or MIN of a group of linear expressions. */
|
86 |
|
|
typedef struct lambda_linear_expression_s
|
87 |
|
|
{
|
88 |
|
|
lambda_vector coefficients;
|
89 |
|
|
int constant;
|
90 |
|
|
lambda_vector invariant_coefficients;
|
91 |
|
|
int denominator;
|
92 |
|
|
struct lambda_linear_expression_s *next;
|
93 |
|
|
} *lambda_linear_expression;
|
94 |
|
|
|
95 |
|
|
#define LLE_COEFFICIENTS(T) ((T)->coefficients)
|
96 |
|
|
#define LLE_CONSTANT(T) ((T)->constant)
|
97 |
|
|
#define LLE_INVARIANT_COEFFICIENTS(T) ((T)->invariant_coefficients)
|
98 |
|
|
#define LLE_DENOMINATOR(T) ((T)->denominator)
|
99 |
|
|
#define LLE_NEXT(T) ((T)->next)
|
100 |
|
|
|
101 |
|
|
lambda_linear_expression lambda_linear_expression_new (int, int);
|
102 |
|
|
void print_lambda_linear_expression (FILE *, lambda_linear_expression, int,
|
103 |
|
|
int, char);
|
104 |
|
|
|
105 |
|
|
/* Loop structure. Our loop structure consists of a constant representing the
|
106 |
|
|
STEP of the loop, a set of linear expressions representing the LOWER_BOUND
|
107 |
|
|
of the loop, a set of linear expressions representing the UPPER_BOUND of
|
108 |
|
|
the loop, and a set of linear expressions representing the LINEAR_OFFSET of
|
109 |
|
|
the loop. The linear offset is a set of linear expressions that are
|
110 |
|
|
applied to *both* the lower bound, and the upper bound. */
|
111 |
|
|
typedef struct lambda_loop_s
|
112 |
|
|
{
|
113 |
|
|
lambda_linear_expression lower_bound;
|
114 |
|
|
lambda_linear_expression upper_bound;
|
115 |
|
|
lambda_linear_expression linear_offset;
|
116 |
|
|
int step;
|
117 |
|
|
} *lambda_loop;
|
118 |
|
|
|
119 |
|
|
#define LL_LOWER_BOUND(T) ((T)->lower_bound)
|
120 |
|
|
#define LL_UPPER_BOUND(T) ((T)->upper_bound)
|
121 |
|
|
#define LL_LINEAR_OFFSET(T) ((T)->linear_offset)
|
122 |
|
|
#define LL_STEP(T) ((T)->step)
|
123 |
|
|
|
124 |
|
|
/* Loop nest structure.
|
125 |
|
|
The loop nest structure consists of a set of loop structures (defined
|
126 |
|
|
above) in LOOPS, along with an integer representing the DEPTH of the loop,
|
127 |
|
|
and an integer representing the number of INVARIANTS in the loop. Both of
|
128 |
|
|
these integers are used to size the associated coefficient vectors in the
|
129 |
|
|
linear expression structures. */
|
130 |
|
|
typedef struct
|
131 |
|
|
{
|
132 |
|
|
lambda_loop *loops;
|
133 |
|
|
int depth;
|
134 |
|
|
int invariants;
|
135 |
|
|
} *lambda_loopnest;
|
136 |
|
|
|
137 |
|
|
#define LN_LOOPS(T) ((T)->loops)
|
138 |
|
|
#define LN_DEPTH(T) ((T)->depth)
|
139 |
|
|
#define LN_INVARIANTS(T) ((T)->invariants)
|
140 |
|
|
|
141 |
|
|
lambda_loopnest lambda_loopnest_new (int, int);
|
142 |
|
|
lambda_loopnest lambda_loopnest_transform (lambda_loopnest, lambda_trans_matrix);
|
143 |
|
|
struct loop;
|
144 |
|
|
struct loops;
|
145 |
|
|
bool perfect_nest_p (struct loop *);
|
146 |
|
|
bool lambda_transform_legal_p (lambda_trans_matrix, int, varray_type);
|
147 |
|
|
void print_lambda_loopnest (FILE *, lambda_loopnest, char);
|
148 |
|
|
|
149 |
|
|
#define lambda_loop_new() (lambda_loop) ggc_alloc_cleared (sizeof (struct lambda_loop_s))
|
150 |
|
|
|
151 |
|
|
void print_lambda_loop (FILE *, lambda_loop, int, int, char);
|
152 |
|
|
|
153 |
|
|
lambda_matrix lambda_matrix_new (int, int);
|
154 |
|
|
|
155 |
|
|
void lambda_matrix_id (lambda_matrix, int);
|
156 |
|
|
bool lambda_matrix_id_p (lambda_matrix, int);
|
157 |
|
|
void lambda_matrix_copy (lambda_matrix, lambda_matrix, int, int);
|
158 |
|
|
void lambda_matrix_negate (lambda_matrix, lambda_matrix, int, int);
|
159 |
|
|
void lambda_matrix_transpose (lambda_matrix, lambda_matrix, int, int);
|
160 |
|
|
void lambda_matrix_add (lambda_matrix, lambda_matrix, lambda_matrix, int,
|
161 |
|
|
int);
|
162 |
|
|
void lambda_matrix_add_mc (lambda_matrix, int, lambda_matrix, int,
|
163 |
|
|
lambda_matrix, int, int);
|
164 |
|
|
void lambda_matrix_mult (lambda_matrix, lambda_matrix, lambda_matrix,
|
165 |
|
|
int, int, int);
|
166 |
|
|
void lambda_matrix_delete_rows (lambda_matrix, int, int, int);
|
167 |
|
|
void lambda_matrix_row_exchange (lambda_matrix, int, int);
|
168 |
|
|
void lambda_matrix_row_add (lambda_matrix, int, int, int, int);
|
169 |
|
|
void lambda_matrix_row_negate (lambda_matrix mat, int, int);
|
170 |
|
|
void lambda_matrix_row_mc (lambda_matrix, int, int, int);
|
171 |
|
|
void lambda_matrix_col_exchange (lambda_matrix, int, int, int);
|
172 |
|
|
void lambda_matrix_col_add (lambda_matrix, int, int, int, int);
|
173 |
|
|
void lambda_matrix_col_negate (lambda_matrix, int, int);
|
174 |
|
|
void lambda_matrix_col_mc (lambda_matrix, int, int, int);
|
175 |
|
|
int lambda_matrix_inverse (lambda_matrix, lambda_matrix, int);
|
176 |
|
|
void lambda_matrix_hermite (lambda_matrix, int, lambda_matrix, lambda_matrix);
|
177 |
|
|
void lambda_matrix_left_hermite (lambda_matrix, int, int, lambda_matrix, lambda_matrix);
|
178 |
|
|
void lambda_matrix_right_hermite (lambda_matrix, int, int, lambda_matrix, lambda_matrix);
|
179 |
|
|
int lambda_matrix_first_nz_vec (lambda_matrix, int, int, int);
|
180 |
|
|
void lambda_matrix_project_to_null (lambda_matrix, int, int, int,
|
181 |
|
|
lambda_vector);
|
182 |
|
|
void print_lambda_matrix (FILE *, lambda_matrix, int, int);
|
183 |
|
|
|
184 |
|
|
lambda_trans_matrix lambda_trans_matrix_new (int, int);
|
185 |
|
|
bool lambda_trans_matrix_nonsingular_p (lambda_trans_matrix);
|
186 |
|
|
bool lambda_trans_matrix_fullrank_p (lambda_trans_matrix);
|
187 |
|
|
int lambda_trans_matrix_rank (lambda_trans_matrix);
|
188 |
|
|
lambda_trans_matrix lambda_trans_matrix_basis (lambda_trans_matrix);
|
189 |
|
|
lambda_trans_matrix lambda_trans_matrix_padding (lambda_trans_matrix);
|
190 |
|
|
lambda_trans_matrix lambda_trans_matrix_inverse (lambda_trans_matrix);
|
191 |
|
|
void print_lambda_trans_matrix (FILE *, lambda_trans_matrix);
|
192 |
|
|
void lambda_matrix_vector_mult (lambda_matrix, int, int, lambda_vector,
|
193 |
|
|
lambda_vector);
|
194 |
|
|
bool lambda_trans_matrix_id_p (lambda_trans_matrix);
|
195 |
|
|
|
196 |
|
|
lambda_body_vector lambda_body_vector_new (int);
|
197 |
|
|
lambda_body_vector lambda_body_vector_compute_new (lambda_trans_matrix,
|
198 |
|
|
lambda_body_vector);
|
199 |
|
|
void print_lambda_body_vector (FILE *, lambda_body_vector);
|
200 |
|
|
lambda_loopnest gcc_loopnest_to_lambda_loopnest (struct loops *,
|
201 |
|
|
struct loop *,
|
202 |
|
|
VEC(tree,heap) **,
|
203 |
|
|
VEC(tree,heap) **,
|
204 |
|
|
bool);
|
205 |
|
|
void lambda_loopnest_to_gcc_loopnest (struct loop *,
|
206 |
|
|
VEC(tree,heap) *, VEC(tree,heap) *,
|
207 |
|
|
lambda_loopnest, lambda_trans_matrix);
|
208 |
|
|
|
209 |
|
|
|
210 |
|
|
static inline void lambda_vector_negate (lambda_vector, lambda_vector, int);
|
211 |
|
|
static inline void lambda_vector_mult_const (lambda_vector, lambda_vector, int, int);
|
212 |
|
|
static inline void lambda_vector_add (lambda_vector, lambda_vector,
|
213 |
|
|
lambda_vector, int);
|
214 |
|
|
static inline void lambda_vector_add_mc (lambda_vector, int, lambda_vector, int,
|
215 |
|
|
lambda_vector, int);
|
216 |
|
|
static inline void lambda_vector_copy (lambda_vector, lambda_vector, int);
|
217 |
|
|
static inline bool lambda_vector_zerop (lambda_vector, int);
|
218 |
|
|
static inline void lambda_vector_clear (lambda_vector, int);
|
219 |
|
|
static inline bool lambda_vector_equal (lambda_vector, lambda_vector, int);
|
220 |
|
|
static inline int lambda_vector_min_nz (lambda_vector, int, int);
|
221 |
|
|
static inline int lambda_vector_first_nz (lambda_vector, int, int);
|
222 |
|
|
static inline void print_lambda_vector (FILE *, lambda_vector, int);
|
223 |
|
|
|
224 |
|
|
/* Allocate a new vector of given SIZE. */
|
225 |
|
|
|
226 |
|
|
static inline lambda_vector
|
227 |
|
|
lambda_vector_new (int size)
|
228 |
|
|
{
|
229 |
|
|
return ggc_alloc_cleared (size * sizeof(int));
|
230 |
|
|
}
|
231 |
|
|
|
232 |
|
|
|
233 |
|
|
|
234 |
|
|
/* Multiply vector VEC1 of length SIZE by a constant CONST1,
|
235 |
|
|
and store the result in VEC2. */
|
236 |
|
|
|
237 |
|
|
static inline void
|
238 |
|
|
lambda_vector_mult_const (lambda_vector vec1, lambda_vector vec2,
|
239 |
|
|
int size, int const1)
|
240 |
|
|
{
|
241 |
|
|
int i;
|
242 |
|
|
|
243 |
|
|
if (const1 == 0)
|
244 |
|
|
lambda_vector_clear (vec2, size);
|
245 |
|
|
else
|
246 |
|
|
for (i = 0; i < size; i++)
|
247 |
|
|
vec2[i] = const1 * vec1[i];
|
248 |
|
|
}
|
249 |
|
|
|
250 |
|
|
/* Negate vector VEC1 with length SIZE and store it in VEC2. */
|
251 |
|
|
|
252 |
|
|
static inline void
|
253 |
|
|
lambda_vector_negate (lambda_vector vec1, lambda_vector vec2,
|
254 |
|
|
int size)
|
255 |
|
|
{
|
256 |
|
|
lambda_vector_mult_const (vec1, vec2, size, -1);
|
257 |
|
|
}
|
258 |
|
|
|
259 |
|
|
/* VEC3 = VEC1+VEC2, where all three the vectors are of length SIZE. */
|
260 |
|
|
|
261 |
|
|
static inline void
|
262 |
|
|
lambda_vector_add (lambda_vector vec1, lambda_vector vec2,
|
263 |
|
|
lambda_vector vec3, int size)
|
264 |
|
|
{
|
265 |
|
|
int i;
|
266 |
|
|
for (i = 0; i < size; i++)
|
267 |
|
|
vec3[i] = vec1[i] + vec2[i];
|
268 |
|
|
}
|
269 |
|
|
|
270 |
|
|
/* VEC3 = CONSTANT1*VEC1 + CONSTANT2*VEC2. All vectors have length SIZE. */
|
271 |
|
|
|
272 |
|
|
static inline void
|
273 |
|
|
lambda_vector_add_mc (lambda_vector vec1, int const1,
|
274 |
|
|
lambda_vector vec2, int const2,
|
275 |
|
|
lambda_vector vec3, int size)
|
276 |
|
|
{
|
277 |
|
|
int i;
|
278 |
|
|
for (i = 0; i < size; i++)
|
279 |
|
|
vec3[i] = const1 * vec1[i] + const2 * vec2[i];
|
280 |
|
|
}
|
281 |
|
|
|
282 |
|
|
/* Copy the elements of vector VEC1 with length SIZE to VEC2. */
|
283 |
|
|
|
284 |
|
|
static inline void
|
285 |
|
|
lambda_vector_copy (lambda_vector vec1, lambda_vector vec2,
|
286 |
|
|
int size)
|
287 |
|
|
{
|
288 |
|
|
memcpy (vec2, vec1, size * sizeof (*vec1));
|
289 |
|
|
}
|
290 |
|
|
|
291 |
|
|
/* Return true if vector VEC1 of length SIZE is the zero vector. */
|
292 |
|
|
|
293 |
|
|
static inline bool
|
294 |
|
|
lambda_vector_zerop (lambda_vector vec1, int size)
|
295 |
|
|
{
|
296 |
|
|
int i;
|
297 |
|
|
for (i = 0; i < size; i++)
|
298 |
|
|
if (vec1[i] != 0)
|
299 |
|
|
return false;
|
300 |
|
|
return true;
|
301 |
|
|
}
|
302 |
|
|
|
303 |
|
|
/* Clear out vector VEC1 of length SIZE. */
|
304 |
|
|
|
305 |
|
|
static inline void
|
306 |
|
|
lambda_vector_clear (lambda_vector vec1, int size)
|
307 |
|
|
{
|
308 |
|
|
memset (vec1, 0, size * sizeof (*vec1));
|
309 |
|
|
}
|
310 |
|
|
|
311 |
|
|
/* Return true if two vectors are equal. */
|
312 |
|
|
|
313 |
|
|
static inline bool
|
314 |
|
|
lambda_vector_equal (lambda_vector vec1, lambda_vector vec2, int size)
|
315 |
|
|
{
|
316 |
|
|
int i;
|
317 |
|
|
for (i = 0; i < size; i++)
|
318 |
|
|
if (vec1[i] != vec2[i])
|
319 |
|
|
return false;
|
320 |
|
|
return true;
|
321 |
|
|
}
|
322 |
|
|
|
323 |
|
|
/* Return the minimum nonzero element in vector VEC1 between START and N.
|
324 |
|
|
We must have START <= N. */
|
325 |
|
|
|
326 |
|
|
static inline int
|
327 |
|
|
lambda_vector_min_nz (lambda_vector vec1, int n, int start)
|
328 |
|
|
{
|
329 |
|
|
int j;
|
330 |
|
|
int min = -1;
|
331 |
|
|
|
332 |
|
|
gcc_assert (start <= n);
|
333 |
|
|
for (j = start; j < n; j++)
|
334 |
|
|
{
|
335 |
|
|
if (vec1[j])
|
336 |
|
|
if (min < 0 || vec1[j] < vec1[min])
|
337 |
|
|
min = j;
|
338 |
|
|
}
|
339 |
|
|
gcc_assert (min >= 0);
|
340 |
|
|
|
341 |
|
|
return min;
|
342 |
|
|
}
|
343 |
|
|
|
344 |
|
|
/* Return the first nonzero element of vector VEC1 between START and N.
|
345 |
|
|
We must have START <= N. Returns N if VEC1 is the zero vector. */
|
346 |
|
|
|
347 |
|
|
static inline int
|
348 |
|
|
lambda_vector_first_nz (lambda_vector vec1, int n, int start)
|
349 |
|
|
{
|
350 |
|
|
int j = start;
|
351 |
|
|
while (j < n && vec1[j] == 0)
|
352 |
|
|
j++;
|
353 |
|
|
return j;
|
354 |
|
|
}
|
355 |
|
|
|
356 |
|
|
|
357 |
|
|
/* Multiply a vector by a matrix. */
|
358 |
|
|
|
359 |
|
|
static inline void
|
360 |
|
|
lambda_vector_matrix_mult (lambda_vector vect, int m, lambda_matrix mat,
|
361 |
|
|
int n, lambda_vector dest)
|
362 |
|
|
{
|
363 |
|
|
int i, j;
|
364 |
|
|
lambda_vector_clear (dest, n);
|
365 |
|
|
for (i = 0; i < n; i++)
|
366 |
|
|
for (j = 0; j < m; j++)
|
367 |
|
|
dest[i] += mat[j][i] * vect[j];
|
368 |
|
|
}
|
369 |
|
|
|
370 |
|
|
|
371 |
|
|
/* Print out a vector VEC of length N to OUTFILE. */
|
372 |
|
|
|
373 |
|
|
static inline void
|
374 |
|
|
print_lambda_vector (FILE * outfile, lambda_vector vector, int n)
|
375 |
|
|
{
|
376 |
|
|
int i;
|
377 |
|
|
|
378 |
|
|
for (i = 0; i < n; i++)
|
379 |
|
|
fprintf (outfile, "%3d ", vector[i]);
|
380 |
|
|
fprintf (outfile, "\n");
|
381 |
|
|
}
|
382 |
|
|
|
383 |
|
|
/* Returns true when the vector V is lexicographically positive, in
|
384 |
|
|
other words, when the first nonzero element is positive. */
|
385 |
|
|
|
386 |
|
|
static inline bool
|
387 |
|
|
lambda_vector_lexico_pos (lambda_vector v,
|
388 |
|
|
unsigned n)
|
389 |
|
|
{
|
390 |
|
|
unsigned i;
|
391 |
|
|
for (i = 0; i < n; i++)
|
392 |
|
|
{
|
393 |
|
|
if (v[i] == 0)
|
394 |
|
|
continue;
|
395 |
|
|
if (v[i] < 0)
|
396 |
|
|
return false;
|
397 |
|
|
if (v[i] > 0)
|
398 |
|
|
return true;
|
399 |
|
|
}
|
400 |
|
|
return true;
|
401 |
|
|
}
|
402 |
|
|
|
403 |
|
|
#endif /* LAMBDA_H */
|
404 |
|
|
|