From d33cec18a9c0d9382dbf1e2b629f3daef29226ee Mon Sep 17 00:00:00 2001 From: Evan Cheng Date: Fri, 20 Feb 2009 22:16:49 +0000 Subject: [PATCH] Fix strange logic in CollectIVUsers used to determine whether all uses are addresses, part 1. This fixes an obvious logic bug. Previously if the only in-loop use is a PHI, it would return AllUsesAreAddresses as true. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@65178 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Transforms/Scalar/LoopStrengthReduce.cpp | 9 +++- test/CodeGen/X86/loop-strength-reduce7.ll | 44 ++++++++++++++++++++ 2 files changed, 52 insertions(+), 1 deletion(-) create mode 100644 test/CodeGen/X86/loop-strength-reduce7.ll diff --git a/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/lib/Transforms/Scalar/LoopStrengthReduce.cpp index a52b7d0f1c2..d18a008fef0 100644 --- a/lib/Transforms/Scalar/LoopStrengthReduce.cpp +++ b/lib/Transforms/Scalar/LoopStrengthReduce.cpp @@ -1439,6 +1439,7 @@ SCEVHandle LoopStrengthReduce::CollectIVUsers(const SCEVHandle &Stride, // fields of the BasedUsers. We do this so that it increases the commonality // of the remaining uses. unsigned NumPHI = 0; + bool HasAddress = false; for (unsigned i = 0, e = UsersToProcess.size(); i != e; ++i) { // If the user is not in the current loop, this means it is using the exit // value of the IV. Do not put anything in the base, make sure it's all in @@ -1449,7 +1450,6 @@ SCEVHandle LoopStrengthReduce::CollectIVUsers(const SCEVHandle &Stride, UsersToProcess[i].Base = SE->getIntegerSCEV(0, UsersToProcess[i].Base->getType()); } else { - // Addressing modes can be folded into loads and stores. Be careful that // the store is through the expression, not of the expression though. bool isPHI = false; @@ -1462,6 +1462,9 @@ SCEVHandle LoopStrengthReduce::CollectIVUsers(const SCEVHandle &Stride, // Not all uses are outside the loop. AllUsesAreOutsideLoop = false; + + if (isAddress) + HasAddress = true; // If this use isn't an address, then not all uses are addresses. if (!isAddress && !isPHI) @@ -1478,6 +1481,10 @@ SCEVHandle LoopStrengthReduce::CollectIVUsers(const SCEVHandle &Stride, if (NumPHI > 1) AllUsesAreAddresses = false; + // There are no in-loop address uses. + if (AllUsesAreAddresses && (!HasAddress && !AllUsesAreOutsideLoop)) + AllUsesAreAddresses = false; + return CommonExprs; } diff --git a/test/CodeGen/X86/loop-strength-reduce7.ll b/test/CodeGen/X86/loop-strength-reduce7.ll new file mode 100644 index 00000000000..b6a130a8619 --- /dev/null +++ b/test/CodeGen/X86/loop-strength-reduce7.ll @@ -0,0 +1,44 @@ +; RUN: llvm-as < %s | llc -march=x86 | not grep imul + +target triple = "i386-apple-darwin9.6" + %struct.III_psy_xmin = type { [22 x double], [13 x [3 x double]] } + %struct.III_scalefac_t = type { [22 x i32], [13 x [3 x i32]] } + %struct.gr_info = type { i32, i32, i32, i32, i32, i32, i32, i32, [3 x i32], [3 x i32], i32, i32, i32, i32, i32, i32, i32, i32, i32, i32*, [4 x i32] } + %struct.lame_global_flags = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i8*, i8*, i32, i32, float, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, float, i32, i32, i32, float, float, float, float, i32, i32, i32, i32, i32, i32, i32, i32 } + +define fastcc void @outer_loop(%struct.lame_global_flags* nocapture %gfp, double* nocapture %xr, i32 %targ_bits, double* nocapture %best_noise, %struct.III_psy_xmin* nocapture %l3_xmin, i32* nocapture %l3_enc, %struct.III_scalefac_t* nocapture %scalefac, %struct.gr_info* nocapture %cod_info, i32 %ch) nounwind { +entry: + br label %bb4 + +bb4: ; preds = %bb4, %entry + br i1 true, label %bb5, label %bb4 + +bb5: ; preds = %bb4 + br i1 true, label %bb28.i37, label %bb.i4 + +bb.i4: ; preds = %bb.i4, %bb5 + br label %bb.i4 + +bb28.i37: ; preds = %bb33.i47, %bb5 + %i.1.reg2mem.0.i = phi i32 [ %0, %bb33.i47 ], [ 0, %bb5 ] ; [#uses=2] + %0 = add i32 %i.1.reg2mem.0.i, 1 ; [#uses=2] + br label %bb29.i38 + +bb29.i38: ; preds = %bb33.i47, %bb28.i37 + %indvar32.i = phi i32 [ %indvar.next33.i, %bb33.i47 ], [ 0, %bb28.i37 ] ; [#uses=2] + %sfb.314.i = add i32 %indvar32.i, 0 ; [#uses=3] + %1 = getelementptr [4 x [21 x double]]* null, i32 0, i32 %0, i32 %sfb.314.i ; [#uses=1] + %2 = load double* %1, align 8 ; [#uses=0] + br i1 false, label %bb30.i41, label %bb33.i47 + +bb30.i41: ; preds = %bb29.i38 + %3 = getelementptr %struct.III_scalefac_t* null, i32 0, i32 1, i32 %sfb.314.i, i32 %i.1.reg2mem.0.i ; [#uses=1] + store i32 0, i32* %3, align 4 + br label %bb33.i47 + +bb33.i47: ; preds = %bb30.i41, %bb29.i38 + %4 = add i32 %sfb.314.i, 1 ; [#uses=1] + %phitmp.i46 = icmp ugt i32 %4, 11 ; [#uses=1] + %indvar.next33.i = add i32 %indvar32.i, 1 ; [#uses=1] + br i1 %phitmp.i46, label %bb28.i37, label %bb29.i38 +}