aboutsummaryrefslogtreecommitdiff
path: root/gcc/tree-ssa-forwprop.c
diff options
context:
space:
mode:
authorJeff Law <law@redhat.com>2005-05-17 16:28:07 +0000
committerJeff Law <law@redhat.com>2005-05-17 16:28:07 +0000
commit42f7e006ba8d6a8e5431cbe7693bd97d804aba42 (patch)
tree86002af17658fce2dd1e335b8fc562dc574e089f /gcc/tree-ssa-forwprop.c
parent93652306fec1f16b5af0d76d2109d2d6715739ee (diff)
* Makefile.in (tree-ssa-forwprop.o): Depend on langhooks.h.
* tree-ssa-forwprop.c: Include langhooks.h. (forward_propagate_addr_expr_into_variable_array_index): New. (forward_propagate_addr_expr): New. (tree_ssa_forward_propagate_single_use_vars): Loop over all the statements in the block instead of just the last statement. Call forward_propagate_addr_expr as needed. (pass_forwprop): Update the SSA graph after forward propagation is complete. * g++.dg/tree-ssa/pr18414.C: New test. * gcc.dg/tree-ssa/pr18414.C: New test. * gcc.dg/tree-ssa/pr17141-1.C: New test. * gcc.dg/tree-ssa/pr17141-2.C: New test. git-svn-id: https://gcc.gnu.org/svn/gcc/trunk@99837 138bc75d-0d04-0410-961f-82ee72b054a4
Diffstat (limited to 'gcc/tree-ssa-forwprop.c')
-rw-r--r--gcc/tree-ssa-forwprop.c304
1 files changed, 294 insertions, 10 deletions
diff --git a/gcc/tree-ssa-forwprop.c b/gcc/tree-ssa-forwprop.c
index 746812ac30d..fb3dcaf6ea8 100644
--- a/gcc/tree-ssa-forwprop.c
+++ b/gcc/tree-ssa-forwprop.c
@@ -1,4 +1,4 @@
-/* Forward propagation of single use variables.
+/* Forward propagation of expressions for single use variables.
Copyright (C) 2004, 2005 Free Software Foundation, Inc.
This file is part of GCC.
@@ -33,13 +33,19 @@ Boston, MA 02111-1307, USA. */
#include "tree-flow.h"
#include "tree-pass.h"
#include "tree-dump.h"
+#include "langhooks.h"
-/* This pass performs simple forward propagation of single use variables
- from their definition site into their single use site.
+/* This pass propagates the RHS of assignment statements into use
+ sites of the LHS of the assignment. It's basically a specialized
+ form of tree combination.
- Right now we only bother forward propagating into COND_EXPRs since those
- are relatively common cases where forward propagation creates valid
- gimple code without the expression needing to fold. i.e.
+ Note carefully that after propagation the resulting statement
+ must still be a proper gimple statement. Right now we simply
+ only perform propagations we know will result in valid gimple
+ code. One day we'll want to generalize this code.
+
+ One class of common cases we handle is forward propagating a single use
+ variale into a COND_EXPR.
bb0:
x = a COND b;
@@ -107,6 +113,37 @@ Boston, MA 02111-1307, USA. */
a statement, we put it back on the worklist to examine on the next
iteration of the main loop.
+ A second class of propagation opportunities arises for ADDR_EXPR
+ nodes.
+
+ ptr = &x->y->z;
+ res = *ptr;
+
+ Will get turned into
+
+ res = x->y->z;
+
+ Or
+
+ ptr = &x[0];
+ ptr2 = ptr + <constant>;
+
+ Will get turned into
+
+ ptr2 = &x[constant/elementsize];
+
+ Or
+
+ ptr = &x[0];
+ offset = index * element_size;
+ offset_p = (pointer) offset;
+ ptr2 = ptr + offset_p
+
+ Will get turned into:
+
+ ptr2 = &x[index];
+
+
This will (of course) be extended as other needs arise. */
/* Given an SSA_NAME VAR, return true if and only if VAR is defined by
@@ -397,6 +434,230 @@ forward_propagate_into_cond (tree cond_expr)
}
}
+/* STMT defines LHS which is contains the address of the 0th element
+ in an array. USE_STMT uses LHS to compute the address of an
+ arbitrary element within the array. The (variable) byte offset
+ of the element is contained in OFFSET.
+
+ We walk back through the use-def chains of OFFSET to verify that
+ it is indeed computing the offset of an element within the array
+ and extract the index corresponding to the given byte offset.
+
+ We then try to fold the entire address expression into a form
+ &array[index].
+
+ If we are successful, we replace the right hand side of USE_STMT
+ with the new address computation. */
+
+static bool
+forward_propagate_addr_into_variable_array_index (tree offset, tree lhs,
+ tree stmt, tree use_stmt)
+{
+ tree index;
+
+ /* The offset must be defined by a simple MODIFY_EXPR statement. */
+ if (TREE_CODE (offset) != MODIFY_EXPR)
+ return false;
+
+ /* The RHS of the statement which defines OFFSET must be a gimple
+ cast of another SSA_NAME. */
+ offset = TREE_OPERAND (offset, 1);
+ if (!is_gimple_cast (offset))
+ return false;
+
+ offset = TREE_OPERAND (offset, 0);
+ if (TREE_CODE (offset) != SSA_NAME)
+ return false;
+
+ /* Get the defining statement of the offset before type
+ conversion. */
+ offset = SSA_NAME_DEF_STMT (offset);
+
+ /* The statement which defines OFFSET before type conversion
+ must be a simple MODIFY_EXPR. */
+ if (TREE_CODE (offset) != MODIFY_EXPR)
+ return false;
+
+ /* The RHS of the statement which defines OFFSET must be a
+ multiplication of an object by the size of the array elements.
+ This implicitly verifies that the size of the array elements
+ is constant. */
+ offset = TREE_OPERAND (offset, 1);
+ if (TREE_CODE (offset) != MULT_EXPR
+ || TREE_CODE (TREE_OPERAND (offset, 1)) != INTEGER_CST
+ || !simple_cst_equal (TREE_OPERAND (offset, 1),
+ TYPE_SIZE_UNIT (TREE_TYPE (TREE_TYPE (lhs)))))
+ return false;
+
+ /* The first operand to the MULT_EXPR is the desired index. */
+ index = TREE_OPERAND (offset, 0);
+
+ /* Replace the pointer addition with array indexing. */
+ TREE_OPERAND (use_stmt, 1) = unshare_expr (TREE_OPERAND (stmt, 1));
+ TREE_OPERAND (TREE_OPERAND (TREE_OPERAND (use_stmt, 1), 0), 1) = index;
+
+ /* That should have created gimple, so there is no need to
+ record information to undo the propagation. */
+ fold_stmt (bsi_stmt_ptr (bsi_for_stmt (use_stmt)));
+ mark_new_vars_to_rename (use_stmt);
+ update_stmt (use_stmt);
+ return true;
+}
+
+/* STMT is a statement of the form SSA_NAME = ADDR_EXPR <whatever>.
+
+ Try to forward propagate the ADDR_EXPR into the uses of the SSA_NAME.
+ Often this will allow for removal of an ADDR_EXPR and INDIRECT_REF
+ node or for recovery of array indexing from pointer arithmetic. */
+
+static bool
+forward_propagate_addr_expr (tree stmt)
+{
+ int stmt_loop_depth = bb_for_stmt (stmt)->loop_depth;
+ tree name = TREE_OPERAND (stmt, 0);
+ use_operand_p imm_use;
+ tree use_stmt, lhs, rhs, array_ref;
+
+ /* We require that the SSA_NAME holding the result of the ADDR_EXPR
+ be used only once. That may be overly conservative in that we
+ could propagate into multiple uses. However, that would effectively
+ be un-cseing the ADDR_EXPR, which is probably not what we want. */
+ single_imm_use (name, &imm_use, &use_stmt);
+ if (!use_stmt)
+ return false;
+
+ /* If the use is not in a simple assignment statement, then
+ there is nothing we can do. */
+ if (TREE_CODE (use_stmt) != MODIFY_EXPR)
+ return false;
+
+ /* If the use is in a deeper loop nest, then we do not want
+ to propagate the ADDR_EXPR into the loop as that is likely
+ adding expression evaluations into the loop. */
+ if (bb_for_stmt (use_stmt)->loop_depth > stmt_loop_depth)
+ return false;
+
+ /* Strip away any outer COMPONENT_REF/ARRAY_REF nodes from the LHS. */
+ lhs = TREE_OPERAND (use_stmt, 0);
+ while (TREE_CODE (lhs) == COMPONENT_REF || TREE_CODE (lhs) == ARRAY_REF)
+ lhs = TREE_OPERAND (lhs, 0);
+
+ /* Now see if the LHS node is an INDIRECT_REF using NAME. If so,
+ propagate the ADDR_EXPR into the use of NAME and fold the result. */
+ if (TREE_CODE (lhs) == INDIRECT_REF && TREE_OPERAND (lhs, 0) == name)
+ {
+ /* This should always succeed in creating gimple, so there is
+ no need to save enough state to undo this propagation. */
+ TREE_OPERAND (lhs, 0) = unshare_expr (TREE_OPERAND (stmt, 1));
+ fold_stmt (bsi_stmt_ptr (bsi_for_stmt (use_stmt)));
+ mark_new_vars_to_rename (use_stmt);
+ update_stmt (use_stmt);
+ return true;
+ }
+
+ /* Trivial case. The use statement could be a trivial copy. We
+ go ahead and handle that case here since it's trivial and
+ removes the need to run copy-prop before this pass to get
+ the best results. Also note that by handling this case here
+ we can catch some cascading effects, ie the single use is
+ in a copy, and the copy is used later by a single INDIRECT_REF
+ for example. */
+ if (TREE_CODE (lhs) == SSA_NAME && TREE_OPERAND (use_stmt, 1) == name)
+ {
+ TREE_OPERAND (use_stmt, 1) = unshare_expr (TREE_OPERAND (stmt, 1));
+ mark_new_vars_to_rename (use_stmt);
+ update_stmt (use_stmt);
+ return true;
+ }
+
+ /* Strip away any outer COMPONENT_REF/ARRAY_REF nodes from the RHS. */
+ rhs = TREE_OPERAND (use_stmt, 1);
+ while (TREE_CODE (rhs) == COMPONENT_REF || TREE_CODE (rhs) == ARRAY_REF)
+ rhs = TREE_OPERAND (rhs, 0);
+
+ /* Now see if the RHS node is an INDIRECT_REF using NAME. If so,
+ propagate the ADDR_EXPR into the use of NAME and fold the result. */
+ if (TREE_CODE (rhs) == INDIRECT_REF && TREE_OPERAND (rhs, 0) == name)
+ {
+ /* This should always succeed in creating gimple, so there is
+ no need to save enough state to undo this propagation. */
+ TREE_OPERAND (rhs, 0) = unshare_expr (TREE_OPERAND (stmt, 1));
+ fold_stmt (bsi_stmt_ptr (bsi_for_stmt (use_stmt)));
+ mark_new_vars_to_rename (use_stmt);
+ update_stmt (use_stmt);
+ return true;
+ }
+
+ /* The remaining cases are all for turning pointer arithmetic into
+ array indexing. They only apply when we have the address of
+ element zero in an array. If that is not the case then there
+ is nothing to do. */
+ array_ref = TREE_OPERAND (TREE_OPERAND (stmt, 1), 0);
+ if (TREE_CODE (array_ref) != ARRAY_REF
+ || TREE_CODE (TREE_TYPE (TREE_OPERAND (array_ref, 0))) != ARRAY_TYPE
+ || !integer_zerop (TREE_OPERAND (array_ref, 1)))
+ return false;
+
+ /* If the use of the ADDR_EXPR must be a PLUS_EXPR, or else there
+ is nothing to do. */
+ if (TREE_CODE (rhs) != PLUS_EXPR)
+ return false;
+
+ /* Try to optimize &x[0] + C where C is a multiple of the size
+ of the elements in X into &x[C/element size]. */
+ if (TREE_OPERAND (rhs, 0) == name
+ && TREE_CODE (TREE_OPERAND (rhs, 1)) == INTEGER_CST)
+ {
+ tree orig = unshare_expr (rhs);
+ TREE_OPERAND (rhs, 0) = unshare_expr (TREE_OPERAND (stmt, 1));
+
+ /* If folding succeeds, then we have just exposed new variables
+ in USE_STMT which will need to be renamed. If folding fails,
+ then we need to put everything back the way it was. */
+ if (fold_stmt (bsi_stmt_ptr (bsi_for_stmt (use_stmt))))
+ {
+ mark_new_vars_to_rename (use_stmt);
+ update_stmt (use_stmt);
+ return true;
+ }
+ else
+ {
+ TREE_OPERAND (use_stmt, 1) = orig;
+ update_stmt (use_stmt);
+ return false;
+ }
+ }
+
+ /* Try to optimize &x[0] + OFFSET where OFFSET is defined by
+ converting a multiplication of an index by the size of the
+ array elements, then the result is converted into the proper
+ type for the arithmetic. */
+ if (TREE_OPERAND (rhs, 0) == name
+ && TREE_CODE (TREE_OPERAND (rhs, 1)) == SSA_NAME
+ /* Avoid problems with IVopts creating PLUS_EXPRs with a
+ different type than their operands. */
+ && lang_hooks.types_compatible_p (TREE_TYPE (name), TREE_TYPE (rhs)))
+ {
+ tree offset_stmt = SSA_NAME_DEF_STMT (TREE_OPERAND (rhs, 1));
+ return forward_propagate_addr_into_variable_array_index (offset_stmt, lhs,
+ stmt, use_stmt);
+ }
+
+ /* Same as the previous case, except the operands of the PLUS_EXPR
+ were reversed. */
+ if (TREE_OPERAND (rhs, 1) == name
+ && TREE_CODE (TREE_OPERAND (rhs, 0)) == SSA_NAME
+ /* Avoid problems with IVopts creating PLUS_EXPRs with a
+ different type than their operands. */
+ && lang_hooks.types_compatible_p (TREE_TYPE (name), TREE_TYPE (rhs)))
+ {
+ tree offset_stmt = SSA_NAME_DEF_STMT (TREE_OPERAND (rhs, 0));
+ return forward_propagate_addr_into_variable_array_index (offset_stmt, lhs,
+ stmt, use_stmt);
+ }
+ return false;
+}
+
/* Main entry point for the forward propagation optimizer. */
static void
@@ -406,9 +667,32 @@ tree_ssa_forward_propagate_single_use_vars (void)
FOR_EACH_BB (bb)
{
- tree last = last_stmt (bb);
- if (last && TREE_CODE (last) == COND_EXPR)
- forward_propagate_into_cond (last);
+ block_stmt_iterator bsi;
+
+ /* Note we update BSI within the loop as necessary. */
+ for (bsi = bsi_start (bb); !bsi_end_p (bsi); )
+ {
+ tree stmt = bsi_stmt (bsi);
+
+ /* If this statement sets an SSA_NAME to an address,
+ try to propagate the address into the uses of the SSA_NAME. */
+ if (TREE_CODE (stmt) == MODIFY_EXPR
+ && TREE_CODE (TREE_OPERAND (stmt, 1)) == ADDR_EXPR
+ && TREE_CODE (TREE_OPERAND (stmt, 0)) == SSA_NAME)
+ {
+ if (forward_propagate_addr_expr (stmt))
+ bsi_remove (&bsi);
+ else
+ bsi_next (&bsi);
+ }
+ else if (TREE_CODE (stmt) == COND_EXPR)
+ {
+ forward_propagate_into_cond (stmt);
+ bsi_next (&bsi);
+ }
+ else
+ bsi_next (&bsi);
+ }
}
}
@@ -433,6 +717,6 @@ struct tree_opt_pass pass_forwprop = {
0, /* properties_destroyed */
0, /* todo_flags_start */
TODO_dump_func | TODO_ggc_collect /* todo_flags_finish */
- | TODO_verify_ssa,
+ | TODO_update_ssa | TODO_verify_ssa,
0 /* letter */
};