1 /* Lambda matrix and vector interface.
2 Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
3 Free Software Foundation, Inc.
4 Contributed by Daniel Berlin <dberlin@dberlin.org>
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
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
31 typedef int *lambda_vector
;
32 DEF_VEC_P(lambda_vector
);
33 DEF_VEC_ALLOC_P(lambda_vector
,heap
);
34 DEF_VEC_ALLOC_P(lambda_vector
,gc
);
36 typedef VEC(lambda_vector
, heap
) *lambda_vector_vec_p
;
37 DEF_VEC_P (lambda_vector_vec_p
);
38 DEF_VEC_ALLOC_P (lambda_vector_vec_p
, heap
);
40 /* An integer matrix. A matrix consists of m vectors of length n (IE
41 all vectors are the same length). */
42 typedef lambda_vector
*lambda_matrix
;
44 DEF_VEC_P (lambda_matrix
);
45 DEF_VEC_ALLOC_P (lambda_matrix
, heap
);
47 /* A transformation matrix, which is a self-contained ROWSIZE x COLSIZE
48 matrix. Rather than use floats, we simply keep a single DENOMINATOR that
49 represents the denominator for every element in the matrix. */
50 typedef struct lambda_trans_matrix_s
56 } *lambda_trans_matrix
;
57 #define LTM_MATRIX(T) ((T)->matrix)
58 #define LTM_ROWSIZE(T) ((T)->rowsize)
59 #define LTM_COLSIZE(T) ((T)->colsize)
60 #define LTM_DENOMINATOR(T) ((T)->denominator)
62 /* A vector representing a statement in the body of a loop.
63 The COEFFICIENTS vector contains a coefficient for each induction variable
64 in the loop nest containing the statement.
65 The DENOMINATOR represents the denominator for each coefficient in the
68 This structure is used during code generation in order to rewrite the old
69 induction variable uses in a statement in terms of the newly created
70 induction variables. */
71 typedef struct lambda_body_vector_s
73 lambda_vector coefficients
;
76 } *lambda_body_vector
;
77 #define LBV_COEFFICIENTS(T) ((T)->coefficients)
78 #define LBV_SIZE(T) ((T)->size)
79 #define LBV_DENOMINATOR(T) ((T)->denominator)
81 /* Piecewise linear expression.
82 This structure represents a linear expression with terms for the invariants
83 and induction variables of a loop.
84 COEFFICIENTS is a vector of coefficients for the induction variables, one
85 per loop in the loop nest.
86 CONSTANT is the constant portion of the linear expression
87 INVARIANT_COEFFICIENTS is a vector of coefficients for the loop invariants,
89 DENOMINATOR is the denominator for all of the coefficients and constants in
91 The linear expressions can be linked together using the NEXT field, in
92 order to represent MAX or MIN of a group of linear expressions. */
93 typedef struct lambda_linear_expression_s
95 lambda_vector coefficients
;
97 lambda_vector invariant_coefficients
;
99 struct lambda_linear_expression_s
*next
;
100 } *lambda_linear_expression
;
102 #define LLE_COEFFICIENTS(T) ((T)->coefficients)
103 #define LLE_CONSTANT(T) ((T)->constant)
104 #define LLE_INVARIANT_COEFFICIENTS(T) ((T)->invariant_coefficients)
105 #define LLE_DENOMINATOR(T) ((T)->denominator)
106 #define LLE_NEXT(T) ((T)->next)
110 lambda_linear_expression
lambda_linear_expression_new (int, int,
112 void print_lambda_linear_expression (FILE *, lambda_linear_expression
, int,
115 /* Loop structure. Our loop structure consists of a constant representing the
116 STEP of the loop, a set of linear expressions representing the LOWER_BOUND
117 of the loop, a set of linear expressions representing the UPPER_BOUND of
118 the loop, and a set of linear expressions representing the LINEAR_OFFSET of
119 the loop. The linear offset is a set of linear expressions that are
120 applied to *both* the lower bound, and the upper bound. */
121 typedef struct lambda_loop_s
123 lambda_linear_expression lower_bound
;
124 lambda_linear_expression upper_bound
;
125 lambda_linear_expression linear_offset
;
129 #define LL_LOWER_BOUND(T) ((T)->lower_bound)
130 #define LL_UPPER_BOUND(T) ((T)->upper_bound)
131 #define LL_LINEAR_OFFSET(T) ((T)->linear_offset)
132 #define LL_STEP(T) ((T)->step)
134 /* Loop nest structure.
135 The loop nest structure consists of a set of loop structures (defined
136 above) in LOOPS, along with an integer representing the DEPTH of the loop,
137 and an integer representing the number of INVARIANTS in the loop. Both of
138 these integers are used to size the associated coefficient vectors in the
139 linear expression structures. */
140 typedef struct lambda_loopnest_s
147 #define LN_LOOPS(T) ((T)->loops)
148 #define LN_DEPTH(T) ((T)->depth)
149 #define LN_INVARIANTS(T) ((T)->invariants)
151 lambda_loopnest
lambda_loopnest_new (int, int, struct obstack
*);
152 lambda_loopnest
lambda_loopnest_transform (lambda_loopnest
,
156 bool perfect_nest_p (struct loop
*);
157 void print_lambda_loopnest (FILE *, lambda_loopnest
, char);
159 void print_lambda_loop (FILE *, lambda_loop
, int, int, char);
161 lambda_matrix
lambda_matrix_new (int, int, struct obstack
*);
163 void lambda_matrix_id (lambda_matrix
, int);
164 bool lambda_matrix_id_p (lambda_matrix
, int);
165 void lambda_matrix_copy (lambda_matrix
, lambda_matrix
, int, int);
166 void lambda_matrix_negate (lambda_matrix
, lambda_matrix
, int, int);
167 void lambda_matrix_transpose (lambda_matrix
, lambda_matrix
, int, int);
168 void lambda_matrix_add (lambda_matrix
, lambda_matrix
, lambda_matrix
, int,
170 void lambda_matrix_add_mc (lambda_matrix
, int, lambda_matrix
, int,
171 lambda_matrix
, int, int);
172 void lambda_matrix_mult (lambda_matrix
, lambda_matrix
, lambda_matrix
,
174 void lambda_matrix_delete_rows (lambda_matrix
, int, int, int);
175 void lambda_matrix_row_exchange (lambda_matrix
, int, int);
176 void lambda_matrix_row_add (lambda_matrix
, int, int, int, int);
177 void lambda_matrix_row_negate (lambda_matrix mat
, int, int);
178 void lambda_matrix_row_mc (lambda_matrix
, int, int, int);
179 void lambda_matrix_col_exchange (lambda_matrix
, int, int, int);
180 void lambda_matrix_col_add (lambda_matrix
, int, int, int, int);
181 void lambda_matrix_col_negate (lambda_matrix
, int, int);
182 void lambda_matrix_col_mc (lambda_matrix
, int, int, int);
183 int lambda_matrix_inverse (lambda_matrix
, lambda_matrix
, int, struct obstack
*);
184 void lambda_matrix_hermite (lambda_matrix
, int, lambda_matrix
, lambda_matrix
);
185 void lambda_matrix_left_hermite (lambda_matrix
, int, int, lambda_matrix
, lambda_matrix
);
186 void lambda_matrix_right_hermite (lambda_matrix
, int, int, lambda_matrix
, lambda_matrix
);
187 int lambda_matrix_first_nz_vec (lambda_matrix
, int, int, int);
188 void lambda_matrix_project_to_null (lambda_matrix
, int, int, int,
190 void print_lambda_matrix (FILE *, lambda_matrix
, int, int);
192 lambda_trans_matrix
lambda_trans_matrix_new (int, int, struct obstack
*);
193 bool lambda_trans_matrix_nonsingular_p (lambda_trans_matrix
);
194 bool lambda_trans_matrix_fullrank_p (lambda_trans_matrix
);
195 int lambda_trans_matrix_rank (lambda_trans_matrix
);
196 lambda_trans_matrix
lambda_trans_matrix_basis (lambda_trans_matrix
);
197 lambda_trans_matrix
lambda_trans_matrix_padding (lambda_trans_matrix
);
198 lambda_trans_matrix
lambda_trans_matrix_inverse (lambda_trans_matrix
,
200 void print_lambda_trans_matrix (FILE *, lambda_trans_matrix
);
201 void lambda_matrix_vector_mult (lambda_matrix
, int, int, lambda_vector
,
203 bool lambda_trans_matrix_id_p (lambda_trans_matrix
);
205 lambda_body_vector
lambda_body_vector_new (int, struct obstack
*);
206 lambda_body_vector
lambda_body_vector_compute_new (lambda_trans_matrix
,
209 void print_lambda_body_vector (FILE *, lambda_body_vector
);
210 lambda_loopnest
gcc_loopnest_to_lambda_loopnest (struct loop
*,
214 void lambda_loopnest_to_gcc_loopnest (struct loop
*,
215 VEC(tree
,heap
) *, VEC(tree
,heap
) *,
217 lambda_loopnest
, lambda_trans_matrix
,
219 void remove_iv (gimple
);
220 tree
find_induction_var_from_exit_cond (struct loop
*);
222 static inline void lambda_vector_negate (lambda_vector
, lambda_vector
, int);
223 static inline void lambda_vector_mult_const (lambda_vector
, lambda_vector
, int, int);
224 static inline void lambda_vector_add (lambda_vector
, lambda_vector
,
226 static inline void lambda_vector_add_mc (lambda_vector
, int, lambda_vector
, int,
228 static inline void lambda_vector_copy (lambda_vector
, lambda_vector
, int);
229 static inline bool lambda_vector_zerop (lambda_vector
, int);
230 static inline void lambda_vector_clear (lambda_vector
, int);
231 static inline bool lambda_vector_equal (lambda_vector
, lambda_vector
, int);
232 static inline int lambda_vector_min_nz (lambda_vector
, int, int);
233 static inline int lambda_vector_first_nz (lambda_vector
, int, int);
234 static inline void print_lambda_vector (FILE *, lambda_vector
, int);
236 /* Allocate a new vector of given SIZE. */
238 static inline lambda_vector
239 lambda_vector_new (int size
)
241 return (lambda_vector
) ggc_alloc_cleared_atomic (sizeof (int) * size
);
246 /* Multiply vector VEC1 of length SIZE by a constant CONST1,
247 and store the result in VEC2. */
250 lambda_vector_mult_const (lambda_vector vec1
, lambda_vector vec2
,
251 int size
, int const1
)
256 lambda_vector_clear (vec2
, size
);
258 for (i
= 0; i
< size
; i
++)
259 vec2
[i
] = const1
* vec1
[i
];
262 /* Negate vector VEC1 with length SIZE and store it in VEC2. */
265 lambda_vector_negate (lambda_vector vec1
, lambda_vector vec2
,
268 lambda_vector_mult_const (vec1
, vec2
, size
, -1);
271 /* VEC3 = VEC1+VEC2, where all three the vectors are of length SIZE. */
274 lambda_vector_add (lambda_vector vec1
, lambda_vector vec2
,
275 lambda_vector vec3
, int size
)
278 for (i
= 0; i
< size
; i
++)
279 vec3
[i
] = vec1
[i
] + vec2
[i
];
282 /* VEC3 = CONSTANT1*VEC1 + CONSTANT2*VEC2. All vectors have length SIZE. */
285 lambda_vector_add_mc (lambda_vector vec1
, int const1
,
286 lambda_vector vec2
, int const2
,
287 lambda_vector vec3
, int size
)
290 for (i
= 0; i
< size
; i
++)
291 vec3
[i
] = const1
* vec1
[i
] + const2
* vec2
[i
];
294 /* Copy the elements of vector VEC1 with length SIZE to VEC2. */
297 lambda_vector_copy (lambda_vector vec1
, lambda_vector vec2
,
300 memcpy (vec2
, vec1
, size
* sizeof (*vec1
));
303 /* Return true if vector VEC1 of length SIZE is the zero vector. */
306 lambda_vector_zerop (lambda_vector vec1
, int size
)
309 for (i
= 0; i
< size
; i
++)
315 /* Clear out vector VEC1 of length SIZE. */
318 lambda_vector_clear (lambda_vector vec1
, int size
)
320 memset (vec1
, 0, size
* sizeof (*vec1
));
323 /* Return true if two vectors are equal. */
326 lambda_vector_equal (lambda_vector vec1
, lambda_vector vec2
, int size
)
329 for (i
= 0; i
< size
; i
++)
330 if (vec1
[i
] != vec2
[i
])
335 /* Return the minimum nonzero element in vector VEC1 between START and N.
336 We must have START <= N. */
339 lambda_vector_min_nz (lambda_vector vec1
, int n
, int start
)
344 gcc_assert (start
<= n
);
345 for (j
= start
; j
< n
; j
++)
348 if (min
< 0 || vec1
[j
] < vec1
[min
])
351 gcc_assert (min
>= 0);
356 /* Return the first nonzero element of vector VEC1 between START and N.
357 We must have START <= N. Returns N if VEC1 is the zero vector. */
360 lambda_vector_first_nz (lambda_vector vec1
, int n
, int start
)
363 while (j
< n
&& vec1
[j
] == 0)
369 /* Multiply a vector by a matrix. */
372 lambda_vector_matrix_mult (lambda_vector vect
, int m
, lambda_matrix mat
,
373 int n
, lambda_vector dest
)
376 lambda_vector_clear (dest
, n
);
377 for (i
= 0; i
< n
; i
++)
378 for (j
= 0; j
< m
; j
++)
379 dest
[i
] += mat
[j
][i
] * vect
[j
];
382 /* Compare two vectors returning an integer less than, equal to, or
383 greater than zero if the first argument is considered to be respectively
384 less than, equal to, or greater than the second.
385 We use the lexicographic order. */
388 lambda_vector_compare (lambda_vector vec1
, int length1
, lambda_vector vec2
,
394 if (length1
< length2
)
395 min_length
= length1
;
397 min_length
= length2
;
399 for (i
= 0; i
< min_length
; i
++)
400 if (vec1
[i
] < vec2
[i
])
402 else if (vec1
[i
] > vec2
[i
])
407 return length1
- length2
;
410 /* Print out a vector VEC of length N to OUTFILE. */
413 print_lambda_vector (FILE * outfile
, lambda_vector vector
, int n
)
417 for (i
= 0; i
< n
; i
++)
418 fprintf (outfile
, "%3d ", vector
[i
]);
419 fprintf (outfile
, "\n");
422 /* Compute the greatest common divisor of two numbers using
423 Euclid's algorithm. */
443 /* Compute the greatest common divisor of a VECTOR of SIZE numbers. */
446 lambda_vector_gcd (lambda_vector vector
, int size
)
454 for (i
= 1; i
< size
; i
++)
455 gcd1
= gcd (gcd1
, vector
[i
]);
460 /* Returns true when the vector V is lexicographically positive, in
461 other words, when the first nonzero element is positive. */
464 lambda_vector_lexico_pos (lambda_vector v
,
468 for (i
= 0; i
< n
; i
++)
480 /* Given a vector of induction variables IVS, and a vector of
481 coefficients COEFS, build a tree that is a linear combination of
482 the induction variables. */
485 build_linear_expr (tree type
, lambda_vector coefs
, VEC (tree
, heap
) *ivs
)
489 tree expr
= build_zero_cst (type
);
491 for (i
= 0; VEC_iterate (tree
, ivs
, i
, iv
); i
++)
496 expr
= fold_build2 (PLUS_EXPR
, type
, expr
, iv
);
499 expr
= fold_build2 (PLUS_EXPR
, type
, expr
,
500 fold_build2 (MULT_EXPR
, type
, iv
,
501 build_int_cst (type
, k
)));
507 /* Returns the dependence level for a vector DIST of size LENGTH.
508 LEVEL = 0 means a lexicographic dependence, i.e. a dependence due
509 to the sequence of statements, not carried by any loop. */
512 static inline unsigned
513 dependence_level (lambda_vector dist_vect
, int length
)
517 for (i
= 0; i
< length
; i
++)
518 if (dist_vect
[i
] != 0)
524 #endif /* LAMBDA_H */