similar to: [LLVMdev] Semantics of an Inbounds GetElementPtr

On Sun, May 3, 2015 at 9:57 PM, David Majnemer <david.majnemer at gmail.com> wrote: > > > On Sun, May 3, 2015 at 6:26 PM, Nicholas White <n.j.white at gmail.com> > wrote: > >> Hi - I've got a question about what optimizations the "inbounds" >> keyword of "getelementptr" allows you to use. In the code below, %five >> is loaded

On Mon, May 4, 2015 at 9:40 AM, Nicholas White <n.j.white at gmail.com> wrote: > Thanks - that makes sense. It's interesting that at -O3 the optimizer > can't reduce the below though - I'll dig into it a bit and see if I > can make a patch that fixes it: I'm unsure what you expect to happen below. It's not quite the same testcase. GVN will PRE the loads, so you

Using LLVM ToT and Hal's helpful slide deck [1], I've been trying to use `llvm.assume` to communicate pointer alignment guarantees to vector load and store instructions. For example, in [2] %5 and %9 are guaranteed to be 32-byte aligned. However, if I run this IR through `opt -O3 -datalayout -S`, the vectorized loads and stores are still 1-byte aligned [3]. What's going wrong? Do I

On Mon, May 4, 2015 at 2:19 PM, Nicholas White <n.j.white at gmail.com> wrote: >> It's not quite the same testcase. > Yes - it's an extension of the first test case that I'd expect to be > optimised out in the same way as my earlier example (i.e., store a > value, read it back and branch on it). If you miss out the > "check.first.array.element" block

Hi Duncan, A DragonEgg/GCC-related question: do you know where these strange FRAME tokens originate from (e.g. %struct.FRAME.matmul)? Compiling simple Fortran code with DragonEgg: > cat matmul.f90 subroutine matmul(nx, ny, nz) implicit none integer :: nx, ny, nz real, dimension(nx, ny) :: A real, dimension(ny, nz) :: B real, dimension(nx, nz) :: C integer :: i, j, k real,

Hi, Recently 10% performance regression on an important benchmark showed up after we integrated https://reviews.llvm.org/rL318299. The analysis showed that rL318299 triggered loop rotation on an multi exits loop, and the loop rotation introduced code layout issue. The performance regression is a side-effect of rL318299. I got two testcases a.ll and b.ll attached to illustrate the problem. a.ll

On Mon, Dec 18, 2017 at 5:46 PM Xinliang David Li <davidxl at google.com> wrote: > The introduction of cleanup.cond block in b.ll without loop-rotation > already makes the layout worse than a.ll. > > > Without introducing cleanup.cond block, the layout out is > > entry->while.cond -> while.body->ret > > All the arrows are hot fall through edges which is

Hi all, I'd like to propose an IR extension that allows the inbounds keyword to be attached to indices in a getelementptr constantexpr. By placing the inbounds keyword on an index, any pointer derived from the getelementptr outside of the bounds of the element referred to by that index, other than the pointer one past the end of the element, shall be treated as a poison value. The main

Works for me On Thu, Jan 22, 2015 at 8:27 PM, Daniel Berlin <dberlin at dberlin.org> wrote: > We should use graph edges, so we can do something better at set build time > :) > > > On Thu Jan 22 2015 at 5:20:46 PM George Burgess IV < > george.burgess.iv at gmail.com> wrote: > >> > Should we be added an edge from the inttoptr to all other pointer >>

Hi the appended IR code does not optimize to my liking :) this is the interesting part in x86_64, that got produced via clang -Os: --- movq -16(%r12), %rax movl -4(%rax), %ecx andl $2298949, %ecx ## imm = 0x231445 cmpq $2298949, (%rax,%rcx) ## imm = 0x231445 leaq 8(%rax,%rcx), %rax cmovneq %r15, %rax movl $2298949, %esi ## imm = 0x231445 movq %r12, %rdi movq %r14,

Without the patch is also returns the wrong answer for all of these, it just doesn't cause LICM to promote because it returns PartialAlias (which is still wrong). We return may-alias instead, and now suddenly it's happy to promote them. The broken noalias results exist both before and after my patch: ===== Alias Analysis Evaluator Report ===== 521 Total Alias Queries Performed

Hi Nadav, Arnold, I managed to find some time to work on the pointer reduction, and I got a patch that can make "canVectorize()" pass. Basically what I do is to teach AddReductionVar() about pointers, saying they don't really have an exit instructions, and that (maybe) the final store is a good candidate (is it?). This makes it recognize the writes and reads, but then

No, i mean the actual store instruction looks like "store i16 %conv22, i16* getelementptr inbounds ([16 x i16]* @pA, i64 0, i64 12), align 2, !tbaa !1" Not that the pointer operand comes from a GEP, but it is a constantexpr, whose opcode is GEP. It sucks that there is such a complex thing to be handled as a store operand directly , but such is life ... CFL-AA *should* treat this

When I use clang on an x86-64 to spit out the LLVM, like this clang -O -S -emit-llvm varargstest.c where varargstest.c looks like this int add_em_up(int count, ...) { va_list ap; int i, sum; va_start(ap, count); sum = 0; for (i = 0; i < count; i++) sum += va_arg(ap, int); va_end(ap); return sum; } I see LLVM that looks like it's been customized for the x86-64, versus

To reproduce the issue, please use the command line "opt -simplifycfg filename". target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" target triple = "x86_64-unknown-linux-gnu" %struct.G = type { %struct.ordered_index_node*, i32 } %struct.ordered_index_node = type { %struct.B, %struct.F } %struct.B = type { i32 } %struct.F = type { i32*, i32* }

Hi Johannes, > From the Lang-Ref statement > > "With the inbounds keyword, the result value of the GEP is undefined > if the address is outside the actual underlying allocated object and > not the address one-past-the-end." > > I'd argue that the actual offset value (here 0) is irrelevant. The GEP > value is undefined if inbounds is present and the

Hi all, I have managed to compile libpng using wllvm and obtain the IR of pngpixel ( small tool which is part of libpng ). libpng has a function called png_check_IHDR: void /* PRIVATE */ png_check_IHDR(png_const_structrp png_ptr, png_uint_32 width, png_uint_32 height, int bit_depth, int color_type, int interlace_type, int compression_type, int filter_type) { int error = 0; /*

Hello, I found that GVN doesn't promote identical sequence of instructions in if and else branch to their common predecessors. For example, for the following code snippet pred: … br i1 %cmp, label %if, label %else if: %incdec.ptr.1 = getelementptr inbounds i8, i8* %ptr, i64 1 %cast1 = ptrtoint i8* %incdec.ptr.1 to i64 … else: %incdec.ptr.2 = getelementptr inbounds i8, i8* %ptr,

Hi, The phase of instruction combine cannot handle the instruction %4 = select i1 %tobool.i, metadata !12, metadata !10 generated by the phase of CFG simplification and the compiler generates an assertion failure. I wonder whether this is valid LLVM IR: %4 = select i1 %tobool.i, metadata !12, metadata !10 Before CFGSimplify cond.true.i: ; preds = %entry

Bingo! That works (when coming from C source) Now, I have a serious problem. I am not coming from C but I build the function with the builder. I am also forced to change the signature and load the pointers a,b,c afterwards: define void @bar([8 x i8]* nocapture readonly %arg_ptr) #0 { entrypoint: %0 = bitcast [8 x i8]* %arg_ptr to i32* %1 = load i32* %0, align 4 %2 = getelementptr [8 x