diff options
| author | Dan Gohman <gohman@apple.com> | 2009-05-19 02:15:55 +0000 |
|---|---|---|
| committer | Dan Gohman <gohman@apple.com> | 2009-05-19 02:15:55 +0000 |
| commit | 5be18e84766fb495b0bde3c8244c1df459a18683 (patch) | |
| tree | 77d5bd8b1b5961b4ed2fd11e271e32fe9ce2fd99 /test/Transforms/IndVarSimplify | |
| parent | fb57f1c8ec95714f7eb4650004539e004bb2db02 (diff) | |
| download | llvm-5be18e84766fb495b0bde3c8244c1df459a18683.tar.gz llvm-5be18e84766fb495b0bde3c8244c1df459a18683.tar.bz2 llvm-5be18e84766fb495b0bde3c8244c1df459a18683.tar.xz | |
Teach SCEVExpander to expand arithmetic involving pointers into GEP
instructions. It attempts to create high-level multi-operand GEPs,
though in cases where this isn't possible it falls back to casting
the pointer to i8* and emitting a GEP with that. Using GEP instructions
instead of ptrtoint+arithmetic+inttoptr helps pointer analyses that
don't use ScalarEvolution, such as BasicAliasAnalysis.
Also, make the AddrModeMatcher more aggressive in handling GEPs.
Previously it assumed that operand 0 of a GEP would require a register
in almost all cases. It now does extra checking and can do more
matching if operand 0 of the GEP is foldable. This fixes a problem
that was exposed by SCEVExpander using GEPs.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72093 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/Transforms/IndVarSimplify')
| -rw-r--r-- | test/Transforms/IndVarSimplify/preserve-gep.ll | 39 |
1 files changed, 39 insertions, 0 deletions
diff --git a/test/Transforms/IndVarSimplify/preserve-gep.ll b/test/Transforms/IndVarSimplify/preserve-gep.ll new file mode 100644 index 0000000000..2c8c224fb9 --- /dev/null +++ b/test/Transforms/IndVarSimplify/preserve-gep.ll @@ -0,0 +1,39 @@ +; RUN: llvm-as < %s | opt -indvars | llvm-dis > %t +; RUN: not grep ptrtoint %t +; RUN: not grep inttoptr %t +; RUN: grep getelementptr %t | count 1 + +; Indvars shouldn't leave getelementptrs expanded out as +; inttoptr+ptrtoint in its output in common cases. + +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128" +target triple = "x86_64-unknown-linux-gnu" + %struct.Foo = type { i32, i32, [10 x i32], i32 } + +define void @me(%struct.Foo* nocapture %Bar) nounwind { +entry: + br i1 false, label %return, label %bb.nph + +bb.nph: ; preds = %entry + br label %bb + +bb: ; preds = %bb1, %bb.nph + %i.01 = phi i64 [ %4, %bb1 ], [ 0, %bb.nph ] ; <i64> [#uses=3] + %0 = getelementptr %struct.Foo* %Bar, i64 %i.01, i32 2, i64 3 ; <i32*> [#uses=1] + %1 = load i32* %0, align 4 ; <i32> [#uses=1] + %2 = mul i32 %1, 113 ; <i32> [#uses=1] + %3 = getelementptr %struct.Foo* %Bar, i64 %i.01, i32 2, i64 3 ; <i32*> [#uses=1] + store i32 %2, i32* %3, align 4 + %4 = add i64 %i.01, 1 ; <i64> [#uses=2] + br label %bb1 + +bb1: ; preds = %bb + %phitmp = icmp sgt i64 %4, 19999 ; <i1> [#uses=1] + br i1 %phitmp, label %bb1.return_crit_edge, label %bb + +bb1.return_crit_edge: ; preds = %bb1 + br label %return + +return: ; preds = %bb1.return_crit_edge, %entry + ret void +} |