The goal of this series is to fix the problems present in sparse
when a bitfield in an uninitialized automatic variable is first
set then read-back.
In this case the bitfield itself is initialized but not the
remaining of the structure/word and sparse was not smart
enough to handle it.
It will also solve most infinite loops that can happen
when processing undefined vars.
Changes since v1:
- use the hack from March to work-around the problem
with undefined vars.
- add the missing simplifications to make the
store-and-load-back of a bitfield almost a no-op
as it should be.
Change since v2:
- in patch 7, '-1' really meant 'all ones' but only for the
size of the word, not bigger. Thanks to Linus for noticing it.
- add a note in patch 1 about the infinite loops.
The series is also available in the git repository at:
git://github.com/lucvoo/sparse.git fix-loading-partialy-defined-bitfields-v3
----------------------------------------------------------------
Luc Van Oostenryck (8):
Remove single-store shortcut
new helper: def_opcode()
reuse nbr_pseudo_users()
change the masking when loading bitfields
simplify ((A & M') | B ) & M when M' & M == 0
transform (A & M) >> S to (A >> S) & (M >> S)
transform (A << S) >> S into A & (-1 >> S)
fix: cast of OP_AND only valid if it's an OP_CAST
flow.c | 39 +-----------
linearize.c | 15 +++--
linearize.h | 5 ++
simplify.c | 109 ++++++++++++++++++++++++++++++---
unssa.c | 5 --
validation/bitfield-size.c | 4 +-
validation/optim/store-load-bitfield.c | 66 ++++++++++++++++++++
7 files changed, 183 insertions(+), 60 deletions(-)
create mode 100644 validation/optim/store-load-bitfield.c
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
