[llvm] [ARM] Stop gluing ALU nodes to branches / selects (PR #116970)

[Simon Tatham via llvm-commits]

================
@@ -1965,32 +1965,34 @@ define float @debug_info(float %gamma, float %slopeLimit, i1 %or.cond, double %t
 ; ARM-ENABLE-NEXT:  @ %bb.1: @ %bb3
 ; ARM-ENABLE-NEXT:    push {r4, r7, lr}
 ; ARM-ENABLE-NEXT:    add r7, sp, #4
-; ARM-ENABLE-NEXT:    sub r4, sp, #16
+; ARM-ENABLE-NEXT:    sub r4, sp, #24
----------------
statham-arm wrote:

Any idea what's up with this increased size of stack frame? I can see that it goes with a newly introduced spill and reload of d10 to the `[r4, #16]` slot, but why has this change created the need for an extra spill at all? Has pre-RA scheduling somehow increased pressure?

https://github.com/llvm/llvm-project/pull/116970