diff options
author | Jakob Stoklund Olesen <stoklund@2pi.dk> | 2012-01-24 23:28:42 +0000 |
---|---|---|
committer | Jakob Stoklund Olesen <stoklund@2pi.dk> | 2012-01-24 23:28:42 +0000 |
commit | d36f5af224594719a10d8b190daf4b56cdb7377f (patch) | |
tree | 38419ea63d70da10be2fd74ca7ab8c4a13ebe2e6 /test | |
parent | 1ce6a36610f06a1fb4047ddde1e9bc2f071c0bfb (diff) | |
download | llvm-d36f5af224594719a10d8b190daf4b56cdb7377f.tar.gz llvm-d36f5af224594719a10d8b190daf4b56cdb7377f.tar.bz2 llvm-d36f5af224594719a10d8b190daf4b56cdb7377f.tar.xz |
Set correct <def,undef> flags when lowering REG_SEQUENCE.
A REG_SEQUENCE instruction is lowered into a sequence of partial defs:
%vreg7:ssub_0<def,undef> = COPY %vreg20:ssub_0
%vreg7:ssub_1<def> = COPY %vreg2
%vreg7:ssub_2<def> = COPY %vreg2
%vreg7:ssub_3<def> = COPY %vreg2
The first def needs an <undef> flag to indicate it is the beginning of
the live range, while the other defs are read-modify-write. Previously,
we depended on LiveIntervalAnalysis to notice and fix the missing
<def,undef>, but that solution was never robust, it was causing problems
with ProcessImplicitDefs and the lowering of chained REG_SEQUENCE
instructions.
This fixes PR11841.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@148879 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
-rw-r--r-- | test/CodeGen/ARM/2010-06-29-PartialRedefFastAlloc.ll | 2 | ||||
-rw-r--r-- | test/CodeGen/ARM/2012-01-24-RegSequenceLiveRange.ll | 43 |
2 files changed, 44 insertions, 1 deletions
diff --git a/test/CodeGen/ARM/2010-06-29-PartialRedefFastAlloc.ll b/test/CodeGen/ARM/2010-06-29-PartialRedefFastAlloc.ll index b9d5600d2a..1aee5088ee 100644 --- a/test/CodeGen/ARM/2010-06-29-PartialRedefFastAlloc.ll +++ b/test/CodeGen/ARM/2010-06-29-PartialRedefFastAlloc.ll @@ -12,7 +12,7 @@ target triple = "thumbv7-apple-darwin10" ; CHECK: vld1.64 {d16, d17}, [r{{.}}] ; CHECK-NOT: vld1.64 {d16, d17} -; CHECK: vmov.f64 d19, d16 +; CHECK: vmov.f64 define i32 @test(i8* %arg) nounwind { entry: diff --git a/test/CodeGen/ARM/2012-01-24-RegSequenceLiveRange.ll b/test/CodeGen/ARM/2012-01-24-RegSequenceLiveRange.ll new file mode 100644 index 0000000000..faf65b60be --- /dev/null +++ b/test/CodeGen/ARM/2012-01-24-RegSequenceLiveRange.ll @@ -0,0 +1,43 @@ +; RUN: llc < %s -mcpu=cortex-a8 -verify-machineinstrs -verify-coalescing +; PR11841 +target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64" +target triple = "armv7-none-linux-eabi" + +; This test case is exercising REG_SEQUENCE, and chains of REG_SEQUENCE. +define arm_aapcs_vfpcc void @foo(i8* nocapture %arg, i8* %arg1) nounwind align 2 { +bb: + %tmp = load <2 x float>* undef, align 8, !tbaa !0 + %tmp2 = extractelement <2 x float> %tmp, i32 0 + %tmp3 = insertelement <4 x float> undef, float %tmp2, i32 0 + %tmp4 = insertelement <4 x float> %tmp3, float 0.000000e+00, i32 1 + %tmp5 = insertelement <4 x float> %tmp4, float 0.000000e+00, i32 2 + %tmp6 = insertelement <4 x float> %tmp5, float 0.000000e+00, i32 3 + %tmp7 = extractelement <2 x float> %tmp, i32 1 + %tmp8 = insertelement <4 x float> %tmp3, float %tmp7, i32 1 + %tmp9 = insertelement <4 x float> %tmp8, float 0.000000e+00, i32 2 + %tmp10 = insertelement <4 x float> %tmp9, float 0.000000e+00, i32 3 + %tmp11 = bitcast <4 x float> %tmp6 to <2 x i64> + %tmp12 = shufflevector <2 x i64> %tmp11, <2 x i64> undef, <1 x i32> zeroinitializer + %tmp13 = bitcast <1 x i64> %tmp12 to <2 x float> + %tmp14 = shufflevector <2 x float> %tmp13, <2 x float> undef, <4 x i32> zeroinitializer + %tmp15 = bitcast <4 x float> %tmp14 to <2 x i64> + %tmp16 = shufflevector <2 x i64> %tmp15, <2 x i64> undef, <1 x i32> zeroinitializer + %tmp17 = bitcast <1 x i64> %tmp16 to <2 x float> + %tmp18 = extractelement <2 x float> %tmp17, i32 0 + tail call arm_aapcs_vfpcc void @bar(i8* undef, float %tmp18, float undef, float 0.000000e+00) nounwind + %tmp19 = bitcast <4 x float> %tmp10 to <2 x i64> + %tmp20 = shufflevector <2 x i64> %tmp19, <2 x i64> undef, <1 x i32> zeroinitializer + %tmp21 = bitcast <1 x i64> %tmp20 to <2 x float> + %tmp22 = shufflevector <2 x float> %tmp21, <2 x float> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1> + %tmp23 = bitcast <4 x float> %tmp22 to <2 x i64> + %tmp24 = shufflevector <2 x i64> %tmp23, <2 x i64> undef, <1 x i32> zeroinitializer + %tmp25 = bitcast <1 x i64> %tmp24 to <2 x float> + %tmp26 = extractelement <2 x float> %tmp25, i32 0 + tail call arm_aapcs_vfpcc void @bar(i8* undef, float undef, float %tmp26, float 0.000000e+00) nounwind + ret void +} + +declare arm_aapcs_vfpcc void @bar(i8*, float, float, float) + +!0 = metadata !{metadata !"omnipotent char", metadata !1} +!1 = metadata !{metadata !"Simple C/C++ TBAA", null} |