@@ -121,6 +121,31 @@ static int bb_depends_on_phi(struct basic_block *target, struct basic_block *src
return 0;
}
+///
+// does the BB contains ignorable instructions but a final branch?
+// :note: something could be done for phi-sources but ... we'll see.
+static bool bb_is_forwarder(struct basic_block *bb)
+{
+ struct instruction *insn;
+
+ FOR_EACH_PTR(bb->insns, insn) {
+ if (!insn->bb)
+ continue;
+ switch (insn->opcode) {
+ case OP_NOP:
+ case OP_INLINED_CALL:
+ continue;
+ case OP_CBR:
+ case OP_BR:
+ return true;
+ default:
+ goto out;
+ }
+ } END_FOR_EACH_PTR(insn);
+out:
+ return false;
+}
+
/*
* When we reach here, we have:
* - a basic block that ends in a conditional branch and
@@ -742,6 +767,22 @@ void vrfy_flow(struct entrypoint *ep)
assert(!entry);
}
+static int retarget_parents(struct basic_block *bb, struct basic_block *target)
+{
+ struct basic_block *parent;
+
+ /*
+ * We can't do FOR_EACH_PTR() here, because the parent list
+ * may change when we rewrite the parent.
+ */
+ while ((parent = first_basic_block(bb->parents))) {
+ if (!rewrite_parent_branch(parent, bb, target))
+ return 0;
+ }
+ kill_bb(bb);
+ return REPEAT_CFG_CLEANUP;
+}
+
static void adjust_phisrc(struct basic_block *top, struct instruction *insn)
{
struct instruction *user = get_phinode(insn);
@@ -821,6 +862,43 @@ static int merge_bb(struct basic_block *top, struct basic_block *bot)
return REPEAT_CFG_CLEANUP;
}
+///
+// early simplification of the CFG
+// Three things are done here:
+// # inactive BB are removed
+// # branches to a 'forwarder' BB are redirected to the forwardee.
+// # merge single-child/single-parent BBs.
+int simplify_cfg_early(struct entrypoint *ep)
+{
+ struct basic_block *bb;
+ int changed = 0;
+
+ FOR_EACH_PTR_REVERSE(ep->bbs, bb) {
+ struct instruction *insn;
+ struct basic_block *tgt;
+
+ if (!bb->ep) {
+ DELETE_CURRENT_PTR(bb);
+ changed = REPEAT_CFG_CLEANUP;
+ continue;
+ }
+
+ insn = last_instruction(bb->insns);
+ if (!insn)
+ continue;
+ switch (insn->opcode) {
+ case OP_BR:
+ tgt = insn->bb_true;
+ if (bb_is_forwarder(bb))
+ changed |= retarget_parents(bb, tgt);
+ else if (bb_list_size(tgt->parents) == 1)
+ changed |= merge_bb(bb, tgt);
+ break;
+ }
+ } END_FOR_EACH_PTR_REVERSE(bb);
+ return changed;
+}
+
void pack_basic_blocks(struct entrypoint *ep)
{
struct basic_block *bb;
@@ -18,6 +18,7 @@ extern void kill_dead_stores(struct entrypoint *ep, pseudo_t addr, int local);
extern void simplify_symbol_usage(struct entrypoint *ep);
extern void simplify_memops(struct entrypoint *ep);
extern void pack_basic_blocks(struct entrypoint *ep);
+extern int simplify_cfg_early(struct entrypoint *ep);
extern void convert_instruction_target(struct instruction *insn, pseudo_t src);
extern void remove_dead_insns(struct entrypoint *);
@@ -57,6 +57,11 @@ void optimize(struct entrypoint *ep)
kill_unreachable_bbs(ep);
ir_validate(ep);
+ cfg_postorder(ep);
+ if (simplify_cfg_early(ep))
+ kill_unreachable_bbs(ep);
+ ir_validate(ep);
+
domtree_build(ep);
/*
@@ -28,12 +28,14 @@ foo:
<entry-point>
add.32 %r3 <- %arg1, %arg2
add.32 %r5 <- %r3, $1
+ # call %r6 <- add, %r3, $1
ret.32 %r5
bar:
.L3:
<entry-point>
+ # call %r13 <- add, %r10, $1
ret
@@ -41,6 +43,7 @@ bas:
.L6:
<entry-point>
add.64 %r16 <- "abc", $1
+ # call %r17 <- lstrip, %r14
ret.64 %r16
@@ -48,6 +51,7 @@ qus:
.L9:
<entry-point>
add.64 %r21 <- messg, $1
+ # call %r22 <- lstrip, %r19
ret.64 %r21
@@ -16,6 +16,7 @@ int foo(void)
foo:
.L0:
<entry-point>
+ # call %r1 <- is_const, $42
ret.32 $42
@@ -27,6 +27,7 @@ foo0:
.L0:
<entry-point>
add.32 %r5 <- %arg1, %arg2
+ # call %r6 <- add, %r1, %r2
ret.32 %r5
@@ -34,12 +35,14 @@ foo1:
.L3:
<entry-point>
add.32 %r10 <- %arg1, $1
+ # call %r11 <- add, %r8, $1
ret.32 %r10
foo2:
.L6:
<entry-point>
+ # call %r13 <- add, $1, $2
ret.32 $3
@@ -14,12 +14,12 @@ foo:
.L0:
<entry-point>
seteq.32 %r2 <- %arg1, $3
- cbr %r2, .L1, .L3
+ cbr %r2, .L1, .L2
.L1:
unreachable
-.L3:
+.L2:
ret.32 %arg1
@@ -16,9 +16,9 @@ int foo(int c)
foo:
.L0:
<entry-point>
- cbr %arg1, .L3, .L2
+ cbr %arg1, .L1, .L2
-.L3:
+.L1:
ret.32 $1
.L2:
@@ -13,6 +13,6 @@ int i3(void) { return (***fun)(); } // C99,C11 6.5.3.2p4
*
* check-output-ignore
* check-output-excludes: load
- * check-output-excludes: call
+ * check-output-excludes: \\tcall
* check-output-pattern(5): ret\\..* \\$42
*/
@@ -9,6 +9,6 @@ int foo(int a, int b, int c)
* check-name: cond-expr
* check-command: test-linearize -Wno-decl -fdump-ir=mem2reg $file
* check-output-ignore
- * check-output-pattern(2): phi\\.
- * check-output-pattern(3): phisrc\\.
+ * check-output-pattern(1): phi\\.
+ * check-output-pattern(2): phisrc\\.
*/
@@ -15,7 +15,6 @@ int foo(int p, int q, int a)
* check-output-ignore
* check-output-excludes: load\\.
* check-output-excludes: store\\.
- * check-output-excludes: phi\\..*, .*, .*
- * check-output-pattern(3): phi\\.
- * check-output-pattern(5): phisrc\\.
+ * check-output-pattern(2): phi\\.
+ * check-output-pattern(4): phisrc\\.
*/
@@ -13,6 +13,6 @@ static void foo(void)
* check-command: test-linearize -Wno-decl $file
*
* check-output-ignore
- * check-output-pattern(0,1): phi\\.
+ * check-output-excludes: phi\\.
* check-output-excludes: cbr
*/
The linearization step sometimes creates a lot of intermediate basic blocks, often containing just a branch. Their presence often make things more complicated than needed (more work to do in later phases, visual clutter when inspection the IR 'by hand') and they can sometimes, indirectly hinder some optimizations. Happily, most of them can trivially be optimized away. So, add a CFG simplification phase running very early and doing: *) jump threading (eliminate jump to jump) *) merge single-child/sinle-parents basic blocks. These changes slightly decrease the number of 'context imbalance' warnings (32 less on a total of 995 warnings) and the size of the generated IR (only ~0.4% but this is very significant relatively to most other simplifications). They also seem to improve the kernel tests' running time: before after real 4m19.261s real 4m17.548s user 72m03.634s user 71m34.642s sys 29m05.573s sys 29m01.856s but it's probably just noise. Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com> --- flow.c | 78 ++++++++++++++++++++ flow.h | 1 + optimize.c | 5 ++ validation/call-inlined.c | 4 + validation/expand/builtin_constant_inline0.c | 1 + validation/inline_base0.c | 3 + validation/linear/builtin_unreachable0.c | 4 +- validation/linear/builtin_unreachable1.c | 4 +- validation/linear/call-inline.c | 2 +- validation/mem2reg/cond-expr.c | 4 +- validation/mem2reg/cond-expr5.c | 5 +- validation/optim/cse-size.c | 2 +- 12 files changed, 102 insertions(+), 11 deletions(-)