This patch makes sure FirstCSPop and MBBI never point to DBG_VALUE instructions, which affected the code generated.
Details
Diff Detail
Event Timeline
The change itself looks good to me, but would be nice to simplify/cleanup the testcase.
lib/Target/X86/X86FrameLowering.cpp | ||
---|---|---|
1558–1566 | You could use this to avoid code duplication: if (Opc != Pop32 && Opc != Pop64) break; FirstCSPop = PI; } | |
test/CodeGen/X86/frame-lowering-debug-intrinsic.ll | ||
5 | The test seems bigger than necessary. Some things that could be simplified:
| |
24 | Use CHECK-LABEL noDebug: to improve FileCheck output in case of errors. | |
47 | CHECK-LABEL |
test/CodeGen/X86/frame-lowering-debug-intrinsic.ll | ||
---|---|---|
5 | I've reduced the test case quite a bit. AFAIK the debug data in the test case is the minimum required. |
Thanks, LGTM. Nitpick below.
test/CodeGen/X86/frame-lowering-debug-intrinsic.ll | ||
---|---|---|
3 | Nitpick: I prefer llc -o - %s over llc < %s because the former can be prefixed with lldb -- and still works as expected. |
The LGTM still stands. I often accept a review even with outstanding nitpicks when I trust that the author can fix the remaining items without the need for another review round.
You could use this to avoid code duplication:
`
if (Opc != DBG_VALUE && !Termiantor) {
}
`