to pass around a struct instead of a large set of individual values. This
cleans up the interface and allows more information to be added to the struct
for future targets without requiring changes to each and every target.
NV_CONTRIB
llvm-svn: 157479
When the flag is set to zero, we do not check for errors in malloc_usable_size.
This may be useful to work around a bug in Nvidia drivers prior to 295.*
llvm-svn: 157472
with arbitrary topologies (previously it would give up when hitting a diamond
in the use graph for example). The testcase from PR12764 is now reduced from
a pile of additions to the optimal 1617*%x0+208. In doing this I changed the
previous strategy of dropping all uses for expression leaves to one of dropping
all but one use. This works out more neatly (but required a bunch of tweaks)
and is also safer: some recently fixed bugs during recursive linearization were
because the linearization code thinks it completely owns a node if it has no uses
outside the expression it is linearizing. But if the node was also in another
expression that had been linearized (and thus all uses of the node from that
expression dropped) then the conclusion that it is completely owned by the
expression currently being linearized is wrong. Keeping one use from within each
linearized expression avoids this kind of mistake.
llvm-svn: 157467
-Wsometimes-uninitialized. This detects cases where an explicitly-written branch
inevitably leads to an uninitialized variable use (so either the branch is dead
code or there is an uninitialized use bug).
This chunk of warnings tentatively lives within -Wuninitialized, in order to
give it more visibility to existing Clang users.
llvm-svn: 157458
these functions will end in the sequence
mov %rbp, %rsp
ret
call __stack_chk_fail
instead of the usual mov, ret. The x86 assembly profiler only looked
for functions ending in 'ret' and added the Unwind row describing how to
set the CFA based on that -- the addition of the call insn (which is jumped
to earlier in the function body) threw off that inspection.
Resolves the need to "step" twice to get out of these functions when doing
source-level stepping.
<rdar://problem/11469705>
llvm-svn: 157454
Load custom plugins when running scan-build. This is useful when
additional static analysis Checkers must be provided via clang's plugin
interface.
Loading additional plugins can now be done via the scan-build call:
scan-build -load-plugin <plugin.so>
A patch by Thomas Hauth.
llvm-svn: 157452
Sending async packets can deadlock a program on darwin. We currently allow breakpoint packets and memory read/write packets (for software breakpoints) to be sent while a program is running. In the GDB remote plug-in, we will interrupt the run, send the async packet and resume (currently with the continue packet that caused the program to resume). If the GDB server supports the "vCont" packet, we might have initially continued with each thread stating it should continue. If new threads show up while we are stopped, which happend when running GCD, we can end up with new threads that we aren't mentioning in the continue list. So we start with a thread list of 1,2,3 and continue:
continue thread 1, continue thread 2, continue thread 3
Now we interrupt and set a breakpoint and we actually have threads 1,2,3,4 now when we are about to resume, yet we send:
continue thread 1, continue thread 2, continue thread 3
Any thread that isn't mentioned is currently going to stay suspended. This causes the deadlock.
llvm-svn: 157439
The Hazard checker implements in-order contraints, or interlocked
resources. Ready instructions with hazards do not enter the available
queue and are not visible to other heuristics.
The major code change is the addition of SchedBoundary to encapsulate
the state at the top or bottom of the schedule, including both a
pending and available queue.
The scheduler now counts cycles in sync with the hazard checker. These
are minimum cycle counts based on known hazards.
Targets with no itinerary (x86_64) currently remain at cycle 0. To fix
this, we need to provide some maximum issue width for all targets. We
also need to add the concept of expected latency vs. minimum latency.
llvm-svn: 157427
Greg Clayton