similar to: -msave-args backend support for x86_64

Another solution is to only save the parts of the model object that interest you. As long as they don't include the formula (which is what drags along the environment it was created in), you will save space. E.g., tfun2 <- function(subset) { junk <- 1:1e6 list(subset=subset, lm(Sepal.Length ~ Sepal.Width, data=iris, subset=subset)$coef) } saveSize(tfun2(1:4)) #[1] 152 Bill

In the below, I generate a model from an environment that isn't .GlobalEnv with a large object that is unrelated to the model generation. It seems to save the irrelevant object unnecessarily. In my actual use case, I am running and saving many models in a loop that each use a single large data.frame (that gets collapsed into a small data.frame for estimation), so removing it isn't an

> On 29 Sep 2020, at 09:28, Quentin Colombet <qcolombet at apple.com> wrote: > > + Daniel who added the MCRegister class. > > Ah sorry, I replied too fast. > I mixed up MCPhysReg with MCRegister. > > I was not aware we had such class. > > From a look at it, MCRegister are essentially the same thing as Register. I am guessing that the difference is Register

Yes so long as you're including the invalid space too (IIRC it matters for DBG_VALUE in particular) the reason I didn't do that is that there's a lot more ctors than consumers of MCRegister. It seemed cheaper to do the checks when they're consumed and pretty much every consumer I encountered started with `assert(Reg.isPhysicalRegister() && ...)`. > On 29 Sep 2020, at

> On 29 Sep 2020, at 11:13, Mircea Trofin <mtrofin at google.com> wrote: > > > > On Tue, Sep 29, 2020 at 11:09 AM Daniel Sanders <daniel_l_sanders at apple.com <mailto:daniel_l_sanders at apple.com>> wrote: > Yes so long as you're including the invalid space too (IIRC it matters for DBG_VALUE in particular) the reason I didn't do that is that

On Tue, Sep 29, 2020 at 9:08 AM Quentin Colombet <qcolombet at apple.com> wrote: > Hi, > > Register can represent virtual or physical registers. > MCRegister can only represent physical registers. > That's what I thought, but MCRegister has some stack slot APIs. > Eventually all Register instances are replaced by a MCRegister. > What happens in that case to the

Hi, I'm gettign an assertion fail/crash in X86FrameLowering::GetFrameIndexReference when compiling the following bitcode: https://gist.github.com/carlokok/868cddebeb9acc8ccbac6253de0480b0 I tried removing the llvm.frameaddres calls but that's not it, where can I start looking for what my mistake here is? Code seems to verify just fine. ; #0 0x00e1afe8

Please consider adding the following to the next R release (I understand that it's too late for R-2.1.0). It adds the menu item `Search R Site' in the `Help' menu in Rgui (which calls RSiteSearch() on the input string). Best, Andy --- R-beta/src/gnuwin32/rui.c 2005-03-06 09:41:40.000000000 -0500 +++ R-beta-new/src/gnuwin32/rui.c 2005-04-12 08:21:55.001824500 -0400 @@ -65,7 +65,8 @@

I think r262546 introduced the assumption that allocas are used exactly once with catchpad. It seems easy to fix, though. On Mon, Oct 17, 2016 at 11:51 PM, Carlo Kok via llvm-dev < llvm-dev at lists.llvm.org> wrote: > This turned out to be related to reusing the local used in the catchpad, > for example given the following c++ code: > > extern void rthrow(); > > int

I’ve been experimenting with SEH handling in LLVM, and it seems like the unwind funclets generated by LLVM are much larger than those generated by Microsoft’s CL compiler. I used the following code as a test: void test() { MyClass x; externalFunction(); } Compiling with CL, the unwind funclet that destroys ‘x’ is just two lines of asm: lea rcx, QWORD PTR x$[rdx] jmp ??1MyClass@@QEAA at XZ

Hi, We 've been working on an LLVM backend for High Performance Erlang (HiPE) [1], the native code compiler of Erlang/OTP [2]. ErLLVM [3] targets the X86 and AMD64 architectures for now but there is some ongoing work from a team on the Uppsala University to also support ARM. In our implementation, we have paid special attention on retaining ABI-compatibility with the Erlang Runtime System in

Mingw switched abis with the release of gcc 4.7 (http://gcc.gnu.org/gcc-4.7/changes.html). The main change is that now mingw (like msvc) given thiscall calling convention to methods by default. I think the last bug blocking us to support the new abi has just been fixed. The question now is how to switch. The attached patches simply switch llvm and clang to the new ABI. This is similar to what

Hi list, I am trying to implement the technique outlined in the following paper: http://www.cs.umd.edu/~mwh/papers/martin10ownership.html in LLVM. My approach so far involves the use of an IR level transform (via runOnFunction) to identify memory loads and stores. One thing I need to do (I am pretty sure I need to do it at least) is automatically mark each stack frame as "owned" by the

On Wed, Mar 20, 2013 at 5:21 PM, Timur Iskhodzhanov <timurrrr at google.com>wrote: > 2013/3/20 Timur Iskhodzhanov <timurrrr at google.com>: > > Anton, > > > > [+Eric, Nick, > > the e-mail thread context has been broken a few times, so you should > > probably look at the llvmdev archives. > > It all starts here: > >

Hi, Following Chris' advice, I will rebase the patches and break them in 3 distinct emails (one at a time) in order to be easier for a reviewer to approve/comments. Please note that the three patches while being code-wise independent, they 're strongly-connected *semantically*, meaning that including just a subset of these patches to LLVM's code base is quite weak if the others are

YES, this is the problem! The program work ok, even a 5x larger version works well. Clearly the _chkstk calls must be emitted with ELF target on Windows as well - why not? I'd like to make a patch and fix this right. I experimented with both changes and practically only the lib/Target/X86/X86ISelLowering.cpp fixes the problem. The other change lib/Target/X86/X86FrameLowering.cpp was not

Anton, [+Eric, Nick, the e-mail thread context has been broken a few times, so you should probably look at the llvmdev archives. It all starts here: http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-October/053961.html ] So I've decided to make a new attempt to fix this issue. We've discussed my patches with Eric and came up with the following idea: a) IsTargetWindows should be true for

To say this first: This whole discussion about using virtregs until emit or having growable physregs is hard to argue without actually having experience trying to go either way. Problems when using virtregs throughout the backend until emit time: - The MC layer is using MCPhysReg (which is an uint16_t) and would need retrofitting to support virtregs - VirtRegs are assumed to have a definition,

Hi Susan, Thanks for the clarification, and the test case. I think I know what the problem is now. Saving and restoring RBP is the job of the PEI (PrologEpilogInsertion) pass, which runs after register allocation. To determine which callee-saved physregs actually need to be saved it checks MachineRegisterInfo::isPhysRegOrOverlapUsed(unsigned reg). Your register allocator needs to notify

Hi! > diff --git a/lib/CodeGen/StackSegmenter.cpp b/lib/CodeGen/StackSegmenter.cpp > new file mode 100644 > index 0000000..5ffb8f2 > --- /dev/null > +++ b/lib/CodeGen/StackSegmenter.cpp > @@ -0,0 +1,48 @@ > +//===-- StackSegmenter.h - Prolog/Epilog code insertion -------*- C++ -* --===// > > The comment is obviously incorrect. Thanks. So much for lifting file