similar to: [LLVMdev] a LICM bug (in LLVM-2.7)?

I don't think licm looks at loads/stores to allocas -- these are usually handled by mem2reg which happens much earlier (if you run your example with -mem2reg you'll see it already deleted the store). In fact, licm sinks the stores by converting them to stores to allocas first and running mem2reg on that. If you change your example to void testLICM(int* restrict p) { int i,N=100;

---------- Forwarded message ---------- From: Ryan Taylor <ryta1203 at gmail.com> Date: Mon, Feb 6, 2012 at 10:36 AM Subject: Re: [LLVMdev] Updating PHI for Instruction Domination? To: Eric Christopher <echristo at apple.com> Since I'm not sure which instructions I might want to replicate, would it be possible to cast the Value from "PHINode::getIncomingValue" to

Hi, I am studying SSA and some time ago I asked on this list how to see the phi nodes in the llvm ir output. I learned then to use this command: opt -mem2reg test.ll -S > test_mem2reg.ll However, if you look at the output (attached to this message) there is something I do not understand. At the end of the function @f at line 18, the function returns the variable %a_addr.0. However, this

I guess not since Value is a superclass of Instruction. On Mon, Feb 6, 2012 at 10:36 AM, Ryan Taylor <ryta1203 at gmail.com> wrote: > > > ---------- Forwarded message ---------- > From: Ryan Taylor <ryta1203 at gmail.com> > Date: Mon, Feb 6, 2012 at 10:36 AM > Subject: Re: [LLVMdev] Updating PHI for Instruction Domination? > To: Eric Christopher <echristo at

You're creating a new path that doesn't include L. For all values defined in L and used outside of L, you need to determine the new reaching def. That's specific to your transformation and can't be automated. Once you do that, creating the phi in F is natural. -Andy On Feb 6, 2012, at 11:51 AM, Ryan Taylor <ryta1203 at gmail.com> wrote: > I guess not since Value is a

On May 22, 2014, at 3:51 PM, Chandler Carruth <chandlerc at google.com> wrote: > > On Thu, May 22, 2014 at 4:42 PM, Louis Gerbarg <lgg at apple.com> wrote: > The problem that the above transform is technically illegal because “When indexing into a (optionally packed) structure, only i32 integer constants are allowed (when using a vector of indices they must all be the same

On Fri, Sep 21, 2012 at 3:08 PM, Welson Sun <welson.sun at gmail.com> wrote: > OK, with the restrict type qualifier, it is a little bit better: > > The IR's function signature becomes: > define void @foo(i32* noalias %a, i32* noalias %b, i32* noalias %c) nounwind > { > > Now the AA result: > Function: foo: 13 pointers, 0 call sites > NoAlias: i32* %a,

Hi all, I have problem with ExtractCodeRegion (CodeExtractor.cpp). My original program is as follows. bb: ... %tmp.15 = load %struct.MYSQL_ROWS** %3, align 4 ... bb1: ... %tmp.1 = load %struct.MYSQL_ROWS** %6, align 4 ... bb4: ; preds = %bb1, %bb, %entry %tmp.0 = phi %struct.MYSQL_ROWS* [ null, %entry ], [ %tmp.15, %bb ], [ %tmp.1, %bb1 ]

So my best bet is to try and work in reg2mem mode and then go back to mem2reg? I'm curious, it seems though when you split a block that the phis get updated, right? On Thu, Feb 2, 2012 at 5:05 PM, Eric Christopher <echristo at apple.com> wrote: > Not that I'm aware of. > > -eric > > On Feb 2, 2012, at 3:47 PM, Ryan Taylor wrote: > > So essentially I'm

Here is the result of running mem2reg then basicaa, it is even worse: (%a should be alias to %0, and partial alias to %3) opt -mem2reg -basicaa -aa-eval -print-all-alias-modref-info < foo.s > /dev/null Function: foo: 6 pointers, 0 call sites NoAlias: i32* %a, i32* %b NoAlias: i32* %a, i32* %c NoAlias: i32* %b, i32* %c PartialAlias: i32* %1, i32* %a NoAlias:

I guess I didn't have a clear question. Suppose we have BB1 and BB2 both point to BB3. BB1 has variable x. BB2 also as variable x. BB3 will have PHINode for x with 2 value from BB1 and BB2. BB1 BB2 \ / BB3 Now if BB1 and BB2 is extracted into a function (using ExtractCodeRegion), they will be replaced by a basic block called codeRepl (which has a call to the extracted

Hi all, Let's see an example (from Alexandre Isoard) first: **************************************************************************************** ; RUN: opt -ipsccp -deadargelim -licm -O2 -S < %s @g = internal global i32 0 ; Function Attrs: argmemonly define internal void @foo(i32* nonnull dereferenceable(4) %arg, i32 %val) #0 { entry: store i32 %val, i32* %arg ret void } define

Recently I posted a patch to migrate certain GEPs between basic blocks in cases where doing so would improve the ability of instcombine to merge into more complicated addressing mode (r209049 and r209065). After some build to failures it was rolled back. I now have a patch that no longer causes the regressions I was seeing, but it also no longer can optimize the case I was trying to optimize. As

On Feb 2, 2012, at 5:13 PM, Ryan Taylor wrote: > So my best bet is to try and work in reg2mem mode and then go back to mem2reg? > I wouldn't. That sounds painful. > I'm curious, it seems though when you split a block that the phis get updated, right? Sure. There's code in splitBasicBlock to do this. It should just take a little bit of work to get what you want done.

Hi Maarten, If you look at the start of basic block 2 (bb2) you'll see the following instruction: %a_addr.0 = phi i32 [ 1, %bb ], [ 0, %bb1 ] ; <i32> [#uses=1] This is an SSA phi node which assigns a value of either 1 or 0 to %a_addr.0 depending on whether control reached the PHI node from basic block bb, or bb1. - Lang. On Thu, Sep 23, 2010 at 11:26 PM, maarten faddegon <

Thank you Hal and Stefan! Bug report is filed: https://bugs.llvm.org/show_bug.cgi?id=46717 And Stefan, I think 'attributor' is a really nice pass, and can infer more precise and useful attributes, which may give more opportunities for optimization. But we shouldn't depend on 'attributor' to correct wrong function attributes, because we cannot run 'attributor' after

Hi all, Is there a way to walk through ALL the predecessors of a basic block in a CFG. I tried to iterate over the preds using this method for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++I) { BasicBlock *PredBB = *PI; } but this only gives the immediate predecessors for a basic block. For example, in this sample control flow graph. entry -> bb1 -> bb2 -> bb4

Here is a proposal for improving DbgValueHistoryCalculator and the overall quality of debug locations. Focus: This is about lowering the DBG_VALUE machine instructions to DWARF location lists. Non-focus: This is not about (typical -O0) variables that permanently reside at a frame index and are described with dbg.declare intrinsics in the IR. These variables are stored in the MMI side-table and

On 01/26/2011 07:50 PM, Vu Le wrote: > I guess I didn't have a clear question. > > Suppose we have BB1 and BB2 both point to BB3. > BB1 has variable x. BB2 also as variable x. > BB3 will have PHINode for x with 2 value from BB1 and BB2. > BB1 BB2 > \ / > BB3 > > Now if BB1 and BB2 is extracted into a function > (using ExtractCodeRegion), they

Dear All, Is there a pass (or set of passes) that will replace a phi whose input operands all compute the same value with an instruction that computes that value? In other words, something that will convert: define internal i32 @function(i32 %x) { ... bb1: %y = add %x, 10 ... bb2: %z = add %x, 10 ... bb3: %phi = [bb1, %y], [bb2, %z] into define internal i32 @function(i32 %x) { ... bb1: ...