NPM inliner does not perform alloca merging

[aeubanks]

Bugzilla Link	47909
Version	unspecified
OS	Windows NT
CC	@alinas,@vns-mn,@rnk

Extended Description

The NPM inliner does not perform alloca merging. Lifetimes can help mitigate the issue of not reusing stack slots when inlining multiple functions, but they are not emitted at -O0. If we have something like

void a(void*);

attribute((always_inline)) void b() {
long long l[10000];
a(l);
}

void c() {
b();
b();
b();
b();
b();
b();
b();
b();
b();
b();
b();
b();
b();
}

the stack usage in c() is much much larger under the NPM than in the legacy PM under -O0.

https://godbolt.org/z/bWaWvv

[aeubanks]

The alternative is to use stack coloring. There's even a comment at Inliner.cpp:

/// Flag to disable manual alloca merging.
///
/// Merging of allocas was originally done as a stack-size saving technique
/// prior to LLVM's code generator having support for stack coloring based on
/// lifetime markers. It is now in the process of being removed. To experiment
/// with disabling it and relying fully on lifetime marker based stack
/// coloring, you can pass this flag to LLVM.

However, FastISel and SelectionDAG at -O0 both don't handle lifetime intrinsics. We'd also need to tell clang's added AlwaysInliner at -O0 to emit lifetime intrinsics.

[aeubanks]

Removing this as blocking NPM. NPM + -O0 has been used for a while without too much complaint. -O0 is supposed to be slow/unoptimized. Keeping this bug open for future consideration.

[aeubanks]

nobody has complained in the many years