llvm-6502/test/Transforms/IndVarSimplify/gep-with-mul-base.ll
Dan Gohman a10756ee65 Re-implement the main strength-reduction portion of LoopStrengthReduction.
This new version is much more aggressive about doing "full" reduction in
cases where it reduces register pressure, and also more aggressive about
rewriting induction variables to count down (or up) to zero when doing so
reduces register pressure.

It currently uses fairly simplistic algorithms for finding reuse
opportunities, but it introduces a new framework allows it to combine
multiple strategies at once to form hybrid solutions, instead of doing
all full-reduction or all base+index.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94061 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-21 02:09:26 +00:00

60 lines
2.0 KiB
LLVM

; RUN: opt < %s -indvars -S > %t
; RUN: grep add %t | count 6
; RUN: grep sub %t | count 2
; RUN: grep mul %t | count 6
define void @foo(i64 %n, i64 %m, i64 %o, double* nocapture %p) nounwind {
entry:
%tmp = icmp sgt i64 %n, 0 ; <i1> [#uses=1]
br i1 %tmp, label %bb.nph, label %return
bb.nph: ; preds = %entry
%tmp1 = mul i64 %n, 37 ; <i64> [#uses=1]
%tmp2 = mul i64 %tmp1, %m ; <i64> [#uses=1]
%tmp3 = mul i64 %tmp2, %o ; <i64> [#uses=1]
br label %bb
bb: ; preds = %bb, %bb.nph
%i.01 = phi i64 [ %tmp3, %bb.nph ], [ %tmp13, %bb ] ; <i64> [#uses=3]
%tmp9 = getelementptr double* %p, i64 %i.01 ; <double*> [#uses=1]
%tmp10 = load double* %tmp9, align 8 ; <double> [#uses=1]
%tmp11 = fdiv double %tmp10, 2.100000e+00 ; <double> [#uses=1]
store double %tmp11, double* %tmp9, align 8
%tmp13 = add i64 %i.01, 1 ; <i64> [#uses=2]
%tmp14 = icmp slt i64 %tmp13, %n ; <i1> [#uses=1]
br i1 %tmp14, label %bb, label %return.loopexit
return.loopexit: ; preds = %bb
br label %return
return: ; preds = %return.loopexit, %entry
ret void
}
define void @bar(i64 %n, i64 %m, i64 %o, i64 %q, double* nocapture %p) nounwind {
entry:
%tmp = icmp sgt i64 %n, 0 ; <i1> [#uses=1]
br i1 %tmp, label %bb.nph, label %return
bb.nph: ; preds = %entry
%tmp1 = mul i64 %n, %q ; <i64> [#uses=1]
%tmp2 = mul i64 %tmp1, %m ; <i64> [#uses=1]
%tmp3 = mul i64 %tmp2, %o ; <i64> [#uses=1]
br label %bb
bb: ; preds = %bb, %bb.nph
%i.01 = phi i64 [ %tmp3, %bb.nph ], [ %tmp13, %bb ] ; <i64> [#uses=3]
%tmp9 = getelementptr double* %p, i64 %i.01 ; <double*> [#uses=1]
%tmp10 = load double* %tmp9, align 8 ; <double> [#uses=1]
%tmp11 = fdiv double %tmp10, 2.100000e+00 ; <double> [#uses=1]
store double %tmp11, double* %tmp9, align 8
%tmp13 = add i64 %i.01, 1 ; <i64> [#uses=2]
%tmp14 = icmp slt i64 %tmp13, %n ; <i1> [#uses=1]
br i1 %tmp14, label %bb, label %return.loopexit
return.loopexit: ; preds = %bb
br label %return
return: ; preds = %return.loopexit, %entry
ret void
}