[PEI][PowerPC] Fix false alarm of stack size limit #65559
Conversation
bzEq
requested review from
arsenm,
nickdesaulniers,
chenzheng1030,
nemanjai and
stefanp-ibm
| ; CHECK-NOT: warning: {{.*}} stack frame size ({{.*}}) exceeds limit (4294967295) in function 'large_stack' | ||
| define i8* @large_stack() { | ||
| %s = alloca [281474976710656 x i8], align 1 | ||
| %e = getelementptr i8, i8* %s, i64 0 |
There was a problem hiding this comment.
use opaque pointers ptr
Does it really work for a larger-than-4G limit? Do you not hit any other limit in LLVM codegen that might assume uint32_t or a smaller limit? You probably want to try more more stack operations.
There was a problem hiding this comment.
On PPC64, larger-than-4G stack was supported after 5018a5d.
There was a problem hiding this comment.
And we have some internal users using stack larger than 4G.
| @@ -2740,3 +2740,11 @@ bool PPCFrameLowering::enableShrinkWrapping(const MachineFunction &MF) const { | |||
| return false; | |||
| return !MF.getSubtarget<PPCSubtarget>().is32BitELFABI(); | |||
| } | |||
|
|
|||
| uint64_t PPCFrameLowering::getStackThreshold() const { | |||
| // The scratch register of STUX contains a signed negative 64-bit number. | |||
There was a problem hiding this comment.
I'm not very familiar with ppc, but this comment doesn't mean much to me. Maybe other reviewers with more familiarity can triple check it?
…ack frame allocation and deallocation
There was a problem hiding this comment.
Were we always getting this warning on the tests added in 5018a5d?
My initial reaction to this was that the test case needs to show an actual access on a large stack frame, but we already have that in test/CodeGen/PowerPC/huge-frame-size.ll
Maybe it would be nice to add these cases to the same file since this is very related.
Otherwise, LGTM
I think so, I didn't catch that in |
PPC64 allows stack size up to ((2^63)-1) bytes. Currently llc reports
if the stack allocated is larger than 4G.