https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81698
Bug ID: 81698
Summary: expand_omp_sections generates unusual switch
Product: gcc
Version: 8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libgomp
Assignee: unassigned at gcc dot gnu.org
Reporter: vries at gcc dot gnu.org
CC: jakub at gcc dot gnu.org
Target Milestone: ---
I.
A switch has the default label as its first label:
...
switch (cD.2303) <default: <L2> [INV] [count: INV],
case 1: <L1> [INV] [count: INV],
case 2: <L0> [INV] [count: INV]>
...
And because build_cfg makes edges iterating over the switch labels, the default
edge will have index '0'.
In expand_case, we seem to be relying on this (it seems fragile to me though):
...
edge default_edge = EDGE_SUCC (bb, 0);
...
II.
The switch resulting from expand_omp_sections does not have the default edge as
first edge (in fact, it's the last):
...
switch (.section.0D.1925) <default: <L6> [INV] [count: INV],
case 0: <L0> [INV] [count: INV],
case 1: <L7> [INV] [count: INV],
case 2: <L8> [INV] [count: INV],
case 3: <L9> [INV] [count: INV],
case 4: <L10> [INV] [count: INV],
case 5: <L11> [INV] [count: INV]>
;; succ: 8
;; 7
;; 6
;; 5
;; 4
;; 11
;; 9
;; basic block 8, loop depth 1, freq 0, maybe hot
;; prev block 7, next block 9, flags: (NEW, REACHABLE)
;; pred: 3
<L11>:
# USE = anything
# CLB = anything
barD.1917 (5);
goto <bb 10>; [INV] [count: INV]
;; succ: 10 (FALLTHRU)
;; basic block 9, loop depth 0, freq 0, maybe hot
;; prev block 8, next block 10, flags: (NEW)
;; pred: 3
<L6>:
# USE = anything
# CLB = anything
__builtin_trapD.1087 ();
;; succ:
...
III.
So we need either to:
- fix expand_omp_sections to emit a normal-form switch, or
- stop relying on the first edge being the default edge.
