similar to: [LLVMdev] Bounds checking

On Nov 30, 2007, at 20:59, Jon Harrop wrote: > Does LLVM hoist bounds checks out of inner loops? In practice, LLVM is frequently going to have conservative behaviors which will prevent hoisting the explicit bounds checks your front-end must generate. Specifically, LLVM must use alias analysis to disprove the hypothesis that the array length variable is updated by the body of the loop.

On Nov 30, 2007, at 5:59 PM, Jon Harrop wrote: > > Does LLVM hoist bounds checks out of inner loops? LLVM does loop unswitching, so yes in some cases it does. -Chris

I am just considering various different designs for the code emitted by a HLVM and one alluring approach is to make all local variables immutable and replace loops with (tail) recursive functions. This makes it much easier to reason about injected code such as GC checks. However, I am concerned about the performance of the generated code. Specifically, this is likely to result in small

On Jan 25, 2009, at 12:21 AM, Jon Harrop wrote: > My question is: will LLVM optimize this back into a function > containing a loop > that runs inside a single stack frame that contains the constants > "a" and "x" > or should my compiler perform this rewrite itself? The easiest way to answer this is to write the moral equivalent in C, and see if it happens

On Wednesday 28 January 2009 07:28:38 Chris Lattner wrote: > On Jan 25, 2009, at 12:21 AM, Jon Harrop wrote: > > My question is: will LLVM optimize this back into a function > > containing a loop > > that runs inside a single stack frame that contains the constants > > "a" and "x" > > or should my compiler perform this rewrite itself? > >

John, Sanjoy, Nuno, David, Thanks for the tip, below are the relevant posts from the archives. I am suggesting something similar to Dan's third option below (Tue Nov 29 2011 "the nsw story”, Dan Gohman), when hoisting an instruction with ‘nsw’ it looses the ‘nsw’ attribute, but without saying “add-nsw is a fully side-effecting instruction”. This option was back then seen by

2011/12/14 Rafael Ávila de Espíndola <rafael.espindola at gmail.com>: >> We first perform a speculation transformation, hoisting all of the >> code above the %overflow_check branch: >> >>   %t0 = add nsw i32 %a, %b >>   %t1 = sext i32 %t0 to i64 >>   %t2 = ashr i64 %t1, 31 >>   %t3 = add i64 %t2, 1 >>   %t5 = icmp ult %t3, 2 >>   %t6 =

For doing PRE on the load, it looks like there’s only two things stopping GVN PRE from doing it: * GVN::PerformLoadPRE doesn’t hoist loads that are conditional. Probably this can be overcome with some kind of heuristic that allows it to happen in loops when the blocks where a load would have to be inserted are outside the loop. * IsFullyAvailableInBlock goes around the loop and

Hi, TL;DR: A loop doesn't get vectorized due to the interaction of loop- rotate, licm and instcombine. What to do about it? Full story: In the benchmarks for our out-of-tree target we have a case that we would like to get vectorized, but currently it isn't. I've done some digging to see why and have some kind of idea what prevents it, but I don't know what the best way to fix

Since most add/sub operations compiled from C have the nsw attribute, we cannot simply restrict movement of these instructions. Sure, we could drop nsw when moving these instructions, but there are still many other problems left. Please read more about the topic here: https://blog.regehr.org/archives/1496 For example, doing loop unswitching, followed by inline (just to illustrate that the

Hi, I am debugging private backend and faced interesting problem: sometimes spill hoisting creates double stores. (some output from -debug-only=regalloc). First hoisting: Checking redundant spills for 0 at 16r in %vreg19 [16r,144B:0)[144B,240B:1)[240B,280r:2)[296r,416B:3)[416B,456r:4)[472r,592B:5) 0 at 16r 1 at 144B-phi 2 at 240B-phi 3 at 296r 4 at 416B-phi 5 at 472r Merged to stack int: SS#0

On Mon, Jul 17, 2017 at 5:11 PM, Wei Mi <wmi at google.com> wrote: > On Mon, Jul 17, 2017 at 2:09 PM, Sanjoy Das > <sanjoy at playingwithpointers.com> wrote: >> Hi, >> >> On Mon, Jul 17, 2017 at 1:56 PM, Daniel Berlin via llvm-dev >> <llvm-dev at lists.llvm.org> wrote: >>> >>> >>> On Mon, Jul 17, 2017 at 1:53 PM, Wei Mi

Hi all, I have a question about dereferenceable metadata on load instruction. I have a patch (https://reviews.llvm.org/D31539) for LICM that hoists loads with !invariant.group. The motivation example is devirtualization: struct A { virtual void foo(); }; int bar(); void indirect(A &a) { while(bar()) a.foo(); } With -O2 -fstrict-vtable-pointers we get: define void

Thanks for the tip! I looked into it and it looks like the problem as of SVN HEAD is that the lui and ori instructions in Mips are considered cheap (1-cycle def-use latency) by MachineLICM::IsCheapInstruction(), but are not trivially materializable because their register operands are not always available. This makes MachineLICM::IsProfitableToHoist() return false, preventing the hoist even

On 07/18/2017 06:03 PM, David Majnemer via llvm-dev wrote: > I doubt it is possible for us to try and make any fix which is > predicated on eagerly treating undef in a particular way, refinement > will always cause these problems to come about... > > Given what I've seen in LLVM (and what I've learned from other > compilers), we probably have two choices: > 1.

Hi, On Mon, Jul 17, 2017 at 1:56 PM, Daniel Berlin via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > > On Mon, Jul 17, 2017 at 1:53 PM, Wei Mi <wmi at google.com> wrote: >> >> It seems MemorySSA.cpp is the only real code where we found the >> problem happening. > > > This is doubtful, ¸FWIW :) > >> >> Is it possible to change the

/// Optimizations after all the reg selections and spills are done. void InlineSpiller::postOptimization() { HSpiller.hoistAllSpills(); } Seems a problematic function to me, as hoistAllSpills() uses TII.storeRegToStackSlot() to insert new spills. The problem is, TII.storeRegToStackSlot is allowed to create new virtual registers, which can not be allocated a range as this whole thing is called

PredicateSimplifier is a pretty interesting pass, but it doesn't look like opt invokes it at any standard -Ox level, and so I assume that llvm-gcc also does not use this pass? If that is right, I'm curious about why this is the case -- does it simply not provide enough code speedup to compensate for the increase in compile time? Also, a colleague and I (we both teach advanced

Yes machine-licm can and should hoist constant materialization instructions out of the loop. If it's not doing that, it's probably because the target is not modeling the instruction correctly. I would walk through MachineLICM::IsLoopInvariantInst() in the debugger to figure it out. You can also try compiling the same bitcode for a target like ARM or X86 as a comparison. Evan On Mar 7,

Hi John, John Regehr wrote: > PredicateSimplifier is a pretty interesting pass, but it doesn't look > like opt invokes it at any standard -Ox level, and so I assume that > llvm-gcc also does not use this pass? If that is right, I'm curious > about why this is the case -- does it simply not provide enough code > speedup to compensate for the increase in compile time? I