Displaying 4 results from an estimated 4 matches for "sldi".
Did you mean:
sld
2016 Oct 18
2
A use of RDF to extend register Remat
...node. This def node would be a member of the "oris" statement. Within that
> statement you would then look for use nodes and you'd find the use node for
> R3: "oris r3, *r3*, 35809". From that node, you'd follow the reaching def
> and this would give you: "sldi *r3*, r3, 32". Then you'd look for use nodes
> and you'd get "sldi r3, *r3*, 32". Following this traversal would
> eventually get you to the first "lis".
>
> Getting from use-node to def-node is a single step.
> Visiting other def/use nodes from the s...
2016 Sep 12
6
[RFC] Register Rematerialization (remat) Extension
...tly remat
does
not cover
$ cat ~/tmp/tl.c
void foo(long);
void bar() {
for (int i = 0; i < 1600; ++i)
foo(3494348345984503943);
}
$ clang -O3 -S -o - ~/tmp/tl.c -target powerpc64
...
# BB#0: # %entry
...
lis 3, 12414
ori 3, 3, 27470
sldi 3, 3, 32
oris 3, 3, 35809
ori 30, 3, 20615
...
.LBB0_1: # %for.body
mr 3, 30
bl foo
...
There is a sequence of instructions used to materialize the constant, the
first
one (the lis) is trivially rematerialiable, and the others depend o...
2015 Feb 24
2
[LLVMdev] Question about shouldMergeGEPs in InstructionCombining
On Mon, Feb 23, 2015 at 2:17 PM, Hal Finkel <hfinkel at anl.gov> wrote:
> ----- Original Message -----
> > From: "Francois Pichet" <pichet2000 at gmail.com>
> > To: "LLVM Developers Mailing List" <llvmdev at cs.uiuc.edu>
> > Sent: Sunday, February 22, 2015 5:34:11 PM
> > Subject: [LLVMdev] Question about shouldMergeGEPs in
2015 Feb 25
2
[LLVMdev] Question about shouldMergeGEPs in InstructionCombining
...>
> .LBB0_1: # %for.body
> # =>This Inner Loop Header:
> Depth=1
> slw 8, 5, 4
> ld 9, .LC1 at toc@l(7)
> addi 5, 5, 4
> add 8, 8, 6
> extsw 8, 8
> sldi 8, 8, 2
> lwzx 8, 9, 8
> addi 9, 3, 16
> stw 8, 0(3)
> mr 3, 9
> bdnz .LBB0_1
>
> there are two things wrong here, first:
>
> ld 9, .LC1 at toc@l(7)
>
> this load is loop invariant, and should be hoisted (but was not).
The problem of the non-hoisted TOC lo...