llvm-6502/test/Transforms/LoopStrengthReduce/X86/no_superflous_induction_vars.ll
Arnold Schwaighofer 2ced33808e SCEVExpander: Try hard not to create derived induction variables in other loops
During LSR of one loop we can run into a situation where we have to expand the
start of a recurrence of a loop induction variable in this loop. This start
value is a value derived of the induction variable of a preceeding loop. SCEV
has cannonicalized this value to a different recurrence than the recurrence of
the preceeding loop's induction variable (the type and/or step direction) has
changed). When we come to instantiate this SCEV we created a second induction
variable in this preceeding loop.  This patch tries to base such derived
induction variables of the preceeding loop's induction variable.

This helps twolf on arm and seems to help scimark2 on x86.

Reapply with a fix for the case of a value derived from a pointer.

radar://15970709

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201496 91177308-0d34-0410-b5e6-96231b3b80d8
2014-02-16 15:49:50 +00:00

51 lines
1.5 KiB
LLVM

; RUN: opt -S -loop-reduce -mcpu=corei7-avx -mtriple=x86_64-apple-macosx < %s | FileCheck %s
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
define void @indvar_expansion(i8* nocapture readonly %rowsptr) {
entry:
br label %for.cond
; SCEVExpander used to create induction variables in the loop %for.cond while
; expanding the recurrence start value of loop strength reduced values from
; %vector.body.
; CHECK-LABEL: indvar_expansion
; CHECK: for.cond:
; CHECK-NOT: phi i3
; CHECK: br i1 {{.+}}, label %for.cond
for.cond:
%indvars.iv44 = phi i64 [ %indvars.iv.next45, %for.cond ], [ 0, %entry ]
%cmp = icmp eq i8 undef, 0
%indvars.iv.next45 = add nuw nsw i64 %indvars.iv44, 1
br i1 %cmp, label %for.cond, label %for.cond2
for.cond2:
br i1 undef, label %for.cond2, label %for.body14.lr.ph
for.body14.lr.ph:
%sext = shl i64 %indvars.iv44, 32
%0 = ashr exact i64 %sext, 32
%1 = sub i64 undef, %indvars.iv44
%2 = and i64 %1, 4294967295
%3 = add i64 %2, 1
%fold = add i64 %1, 1
%n.mod.vf = and i64 %fold, 7
%n.vec = sub i64 %3, %n.mod.vf
%end.idx.rnd.down = add i64 %n.vec, %0
br label %vector.body
vector.body:
%index = phi i64 [ %index.next, %vector.body ], [ %0, %for.body14.lr.ph ]
%4 = getelementptr inbounds i8* %rowsptr, i64 %index
%5 = bitcast i8* %4 to <4 x i8>*
%wide.load = load <4 x i8>* %5, align 1
%index.next = add i64 %index, 8
%6 = icmp eq i64 %index.next, %end.idx.rnd.down
br i1 %6, label %for.end24, label %vector.body
for.end24:
ret void
}