Fix for PR14824: "Optimization arm_ldst_opt inserts newly generated instruction vldmia at incorrect position".

Patch introduces memory operands tracking in ARMLoadStoreOpt::LoadStoreMultipleOpti. For each register it keeps the order of load operations as it was before optimization pass.
It is kind of deep improvement of fix proposed by Hao: http://llvm.org/bugs/show_bug.cgi?id=14824#c4
But it also tracks conflicts between different register classes (e.g. D2 and S5).
For more details see:
Bug description: http://llvm.org/bugs/show_bug.cgi?id=14824
LLVM Commits discussion: 
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20130311/167936.html
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20130318/168688.html
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20130325/169376.html
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20130401/170238.html



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@178851 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 110 insertions, 0 deletions

diff --git a/test/CodeGen/ARM/2013-04-05-overridden-loads-PR14824.ll b/test/CodeGen/ARM/2013-04-05-overridden-loads-PR14824.ll
new file mode 100644
index 0000000000..2561686c1f
--- /dev/null
+++ b/test/CodeGen/ARM/2013-04-05-overridden-loads-PR14824.ll
@@ -0,0 +1,110 @@
+; RUN: llc < %s -mtriple=thumbv7-none-linux-gnueabi -mcpu=cortex-a9 -mattr=+neon,+neonfp | FileCheck %s
+; The test is presented by Jiangning Liu.
+;CHECK-NOT: vldmia
+
+define void @sample_test(<8 x i64> * %secondSource, <8 x i64> * %source, <8 x i64> * %dest) nounwind {
+entry:
+  %s0 = load <8 x i64> * %source, align 64
+  %s1 = load <8 x i64> * %secondSource, align 64
+  %s2 = bitcast <8 x i64> %s0 to i512
+  %data.i.i.48.extract.shift = lshr i512 %s2, 384
+  %data.i.i.48.extract.trunc = trunc i512 %data.i.i.48.extract.shift to i64
+  %arrayidx64 = getelementptr inbounds <8 x i64> * %source, i32 6
+  %s120 = load <8 x i64> * %arrayidx64, align 64
+  %arrayidx67 = getelementptr inbounds <8 x i64> * %secondSource, i32 6
+  %s121 = load <8 x i64> * %arrayidx67, align 64
+  %s122 = bitcast <8 x i64> %s120 to i512
+  %data.i.i677.48.extract.shift = lshr i512 %s122, 384
+  %data.i.i677.48.extract.trunc = trunc i512 %data.i.i677.48.extract.shift to i64
+  %s123 = insertelement <8 x i64> undef, i64 %data.i.i677.48.extract.trunc, i32 0
+  %data.i.i677.32.extract.shift = lshr i512 %s122, 256
+  %data.i.i677.32.extract.trunc = trunc i512 %data.i.i677.32.extract.shift to i64
+  %s124 = insertelement <8 x i64> %s123, i64 %data.i.i677.32.extract.trunc, i32 1
+  %data.i.i677.16.extract.shift = lshr i512 %s122, 128
+  %data.i.i677.16.extract.trunc = trunc i512 %data.i.i677.16.extract.shift to i64
+  %s125 = insertelement <8 x i64> %s124, i64 %data.i.i677.16.extract.trunc, i32 2
+  %data.i.i677.56.extract.shift = lshr i512 %s122, 448
+  %data.i.i677.56.extract.trunc = trunc i512 %data.i.i677.56.extract.shift to i64
+  %s126 = insertelement <8 x i64> %s125, i64 %data.i.i677.56.extract.trunc, i32 3
+  %data.i.i677.24.extract.shift = lshr i512 %s122, 192
+  %data.i.i677.24.extract.trunc = trunc i512 %data.i.i677.24.extract.shift to i64
+  %s127 = insertelement <8 x i64> %s126, i64 %data.i.i677.24.extract.trunc, i32 4
+  %s128 = insertelement <8 x i64> %s127, i64 %data.i.i677.32.extract.trunc, i32 5
+  %s129 = insertelement <8 x i64> %s128, i64 %data.i.i677.16.extract.trunc, i32 6
+  %s130 = insertelement <8 x i64> %s129, i64 %data.i.i677.56.extract.trunc, i32 7
+  %s131 = bitcast <8 x i64> %s121 to i512
+  %data.i1.i676.48.extract.shift = lshr i512 %s131, 384
+  %data.i1.i676.48.extract.trunc = trunc i512 %data.i1.i676.48.extract.shift to i64
+  %s132 = insertelement <8 x i64> undef, i64 %data.i1.i676.48.extract.trunc, i32 0
+  %data.i1.i676.32.extract.shift = lshr i512 %s131, 256
+  %data.i1.i676.32.extract.trunc = trunc i512 %data.i1.i676.32.extract.shift to i64
+  %s133 = insertelement <8 x i64> %s132, i64 %data.i1.i676.32.extract.trunc, i32 1
+  %data.i1.i676.16.extract.shift = lshr i512 %s131, 128
+  %data.i1.i676.16.extract.trunc = trunc i512 %data.i1.i676.16.extract.shift to i64
+  %s134 = insertelement <8 x i64> %s133, i64 %data.i1.i676.16.extract.trunc, i32 2
+  %data.i1.i676.56.extract.shift = lshr i512 %s131, 448
+  %data.i1.i676.56.extract.trunc = trunc i512 %data.i1.i676.56.extract.shift to i64
+  %s135 = insertelement <8 x i64> %s134, i64 %data.i1.i676.56.extract.trunc, i32 3
+  %data.i1.i676.24.extract.shift = lshr i512 %s131, 192
+  %data.i1.i676.24.extract.trunc = trunc i512 %data.i1.i676.24.extract.shift to i64
+  %s136 = insertelement <8 x i64> %s135, i64 %data.i1.i676.24.extract.trunc, i32 4
+  %s137 = insertelement <8 x i64> %s136, i64 %data.i1.i676.32.extract.trunc, i32 5
+  %s138 = insertelement <8 x i64> %s137, i64 %data.i1.i676.16.extract.trunc, i32 6
+  %s139 = insertelement <8 x i64> %s138, i64 %data.i1.i676.56.extract.trunc, i32 7
+  %vecinit28.i.i699 = shufflevector <8 x i64> %s139, <8 x i64> %s130, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 undef, i32 undef, i32 undef>
+  %vecinit35.i.i700 = shufflevector <8 x i64> %vecinit28.i.i699, <8 x i64> %s139, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 13, i32 undef, i32 undef>
+  %vecinit42.i.i701 = shufflevector <8 x i64> %vecinit35.i.i700, <8 x i64> %s139, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 14, i32 undef>
+  %vecinit49.i.i702 = shufflevector <8 x i64> %vecinit42.i.i701, <8 x i64> %s130, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 15>
+  %arrayidx72 = getelementptr inbounds <8 x i64> * %dest, i32 6
+  store <8 x i64> %vecinit49.i.i702, <8 x i64> * %arrayidx72, align 64
+  %arrayidx75 = getelementptr inbounds <8 x i64> * %source, i32 7
+  %s140 = load <8 x i64> * %arrayidx75, align 64
+  %arrayidx78 = getelementptr inbounds <8 x i64> * %secondSource, i32 7
+  %s141 = load <8 x i64> * %arrayidx78, align 64
+  %s142 = bitcast <8 x i64> %s140 to i512
+  %data.i.i650.32.extract.shift = lshr i512 %s142, 256
+  %data.i.i650.32.extract.trunc = trunc i512 %data.i.i650.32.extract.shift to i64
+  %s143 = insertelement <8 x i64> undef, i64 %data.i.i650.32.extract.trunc, i32 0
+  %s144 = insertelement <8 x i64> %s143, i64 %data.i.i650.32.extract.trunc, i32 1
+  %data.i.i650.16.extract.shift = lshr i512 %s142, 128
+  %data.i.i650.16.extract.trunc = trunc i512 %data.i.i650.16.extract.shift to i64
+  %s145 = insertelement <8 x i64> %s144, i64 %data.i.i650.16.extract.trunc, i32 2
+  %data.i.i650.8.extract.shift = lshr i512 %s142, 64
+  %data.i.i650.8.extract.trunc = trunc i512 %data.i.i650.8.extract.shift to i64
+  %s146 = insertelement <8 x i64> %s145, i64 %data.i.i650.8.extract.trunc, i32 3
+  %s147 = insertelement <8 x i64> %s146, i64 %data.i.i650.8.extract.trunc, i32 4
+  %data.i.i650.48.extract.shift = lshr i512 %s142, 384
+  %data.i.i650.48.extract.trunc = trunc i512 %data.i.i650.48.extract.shift to i64
+  %s148 = insertelement <8 x i64> %s147, i64 %data.i.i650.48.extract.trunc, i32 5
+  %s149 = insertelement <8 x i64> %s148, i64 %data.i.i650.16.extract.trunc, i32 6
+  %data.i.i650.0.extract.trunc = trunc i512 %s142 to i64
+  %s150 = insertelement <8 x i64> %s149, i64 %data.i.i650.0.extract.trunc, i32 7
+  %s151 = bitcast <8 x i64> %s141 to i512
+  %data.i1.i649.32.extract.shift = lshr i512 %s151, 256
+  %data.i1.i649.32.extract.trunc = trunc i512 %data.i1.i649.32.extract.shift to i64
+  %s152 = insertelement <8 x i64> undef, i64 %data.i1.i649.32.extract.trunc, i32 0
+  %s153 = insertelement <8 x i64> %s152, i64 %data.i1.i649.32.extract.trunc, i32 1
+  %data.i1.i649.16.extract.shift = lshr i512 %s151, 128
+  %data.i1.i649.16.extract.trunc = trunc i512 %data.i1.i649.16.extract.shift to i64
+  %s154 = insertelement <8 x i64> %s153, i64 %data.i1.i649.16.extract.trunc, i32 2
+  %data.i1.i649.8.extract.shift = lshr i512 %s151, 64
+  %data.i1.i649.8.extract.trunc = trunc i512 %data.i1.i649.8.extract.shift to i64
+  %s155 = insertelement <8 x i64> %s154, i64 %data.i1.i649.8.extract.trunc, i32 3
+  %s156 = insertelement <8 x i64> %s155, i64 %data.i1.i649.8.extract.trunc, i32 4
+  %data.i1.i649.48.extract.shift = lshr i512 %s151, 384
+  %data.i1.i649.48.extract.trunc = trunc i512 %data.i1.i649.48.extract.shift to i64
+  %s157 = insertelement <8 x i64> %s156, i64 %data.i1.i649.48.extract.trunc, i32 5
+  %s158 = insertelement <8 x i64> %s157, i64 %data.i1.i649.16.extract.trunc, i32 6
+  %data.i1.i649.0.extract.trunc = trunc i512 %s151 to i64
+  %s159 = insertelement <8 x i64> %s158, i64 %data.i1.i649.0.extract.trunc, i32 7
+  %vecinit7.i.i669 = shufflevector <8 x i64> %s159, <8 x i64> %s150, <8 x i32> <i32 0, i32 9, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %vecinit14.i.i670 = shufflevector <8 x i64> %vecinit7.i.i669, <8 x i64> %s150, <8 x i32> <i32 0, i32 1, i32 10, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %vecinit21.i.i671 = shufflevector <8 x i64> %vecinit14.i.i670, <8 x i64> %s150, <8 x i32> <i32 0, i32 1, i32 2, i32 11, i32 undef, i32 undef, i32 undef, i32 undef>
+  %vecinit28.i.i672 = shufflevector <8 x i64> %vecinit21.i.i671, <8 x i64> %s150, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 undef, i32 undef, i32 undef>
+  %vecinit35.i.i673 = shufflevector <8 x i64> %vecinit28.i.i672, <8 x i64> %s159, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 13, i32 undef, i32 undef>
+  %vecinit42.i.i674 = shufflevector <8 x i64> %vecinit35.i.i673, <8 x i64> %s159, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 14, i32 undef>
+  %vecinit49.i.i675 = shufflevector <8 x i64> %vecinit42.i.i674, <8 x i64> %s159, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 15>
+  %arrayidx83 = getelementptr inbounds <8 x i64> * %dest, i32 7
+  store <8 x i64> %vecinit49.i.i675, <8 x i64> * %arrayidx83, align 64
+  ret void
+}