Teach VirtRegRewriter to handle spilling in instructions that have multiple

definitions of the virtual register.

This happens when spilling the registers produced by REG_SEQUENCE:

%reg1047:5<def>, %reg1047:6<def>, %reg1047:7<def> = VLD3d8 %reg1033, 0, pred:14, pred:%reg0

The rewriter would spill the register multiple times, dead store elimination
tried to keep up, but ended up cutting the branch it was sitting on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104321 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 45 insertions, 0 deletions

diff --git a/test/CodeGen/ARM/2010-05-20-NEONSpillCrash.ll b/test/CodeGen/ARM/2010-05-20-NEONSpillCrash.ll
new file mode 100644
index 00000000000..b6fbf9bdfb1
--- /dev/null
+++ b/test/CodeGen/ARM/2010-05-20-NEONSpillCrash.ll
@@ -0,0 +1,45 @@
+; RUN: llc < %s -march=arm -mattr=+neon -O0
+
+; This test would crash the rewriter when trying to handle a spill after one of
+; the @llvm.arm.neon.vld3.v8i8 defined three parts of a register.
+
+%struct.__neon_int8x8x3_t = type { <8 x i8>, <8 x i8>, <8 x i8> }
+
+declare %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8*) nounwind readonly
+
+declare void @llvm.arm.neon.vst3.v8i8(i8*, <8 x i8>, <8 x i8>, <8 x i8>) nounwind
+
+define <8 x i8> @t3(i8* %A1, i8* %A2, i8* %A3, i8* %A4, i8* %A5, i8* %A6, i8* %A7, i8* %A8, i8* %B) nounwind {
+  %tmp1b = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A2) ; <%struct.__neon_int8x8x3_t> [#uses=2]
+  %tmp2b = extractvalue %struct.__neon_int8x8x3_t %tmp1b, 0 ; <<8 x i8>> [#uses=1]
+  %tmp4b = extractvalue %struct.__neon_int8x8x3_t %tmp1b, 1 ; <<8 x i8>> [#uses=1]
+  %tmp1d = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A4) ; <%struct.__neon_int8x8x3_t> [#uses=2]
+  %tmp2d = extractvalue %struct.__neon_int8x8x3_t %tmp1d, 0 ; <<8 x i8>> [#uses=1]
+  %tmp4d = extractvalue %struct.__neon_int8x8x3_t %tmp1d, 1 ; <<8 x i8>> [#uses=1]
+  %tmp1e = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A5) ; <%struct.__neon_int8x8x3_t> [#uses=1]
+  %tmp2e = extractvalue %struct.__neon_int8x8x3_t %tmp1e, 0 ; <<8 x i8>> [#uses=1]
+  %tmp1f = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A6) ; <%struct.__neon_int8x8x3_t> [#uses=1]
+  %tmp2f = extractvalue %struct.__neon_int8x8x3_t %tmp1f, 0 ; <<8 x i8>> [#uses=1]
+  %tmp1g = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A7) ; <%struct.__neon_int8x8x3_t> [#uses=2]
+  %tmp2g = extractvalue %struct.__neon_int8x8x3_t %tmp1g, 0 ; <<8 x i8>> [#uses=1]
+  %tmp4g = extractvalue %struct.__neon_int8x8x3_t %tmp1g, 1 ; <<8 x i8>> [#uses=1]
+  %tmp1h = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A8) ; <%struct.__neon_int8x8x3_t> [#uses=2]
+  %tmp2h = extractvalue %struct.__neon_int8x8x3_t %tmp1h, 0 ; <<8 x i8>> [#uses=1]
+  %tmp3h = extractvalue %struct.__neon_int8x8x3_t %tmp1h, 2 ; <<8 x i8>> [#uses=1]
+  %tmp2bd = add <8 x i8> %tmp2b, %tmp2d           ; <<8 x i8>> [#uses=1]
+  %tmp4bd = add <8 x i8> %tmp4b, %tmp4d           ; <<8 x i8>> [#uses=1]
+  %tmp2abcd = mul <8 x i8> undef, %tmp2bd         ; <<8 x i8>> [#uses=1]
+  %tmp4abcd = mul <8 x i8> undef, %tmp4bd         ; <<8 x i8>> [#uses=2]
+  call void @llvm.arm.neon.vst3.v8i8(i8* %A1, <8 x i8> %tmp4abcd, <8 x i8> zeroinitializer, <8 x i8> %tmp2abcd)
+  %tmp2ef = sub <8 x i8> %tmp2e, %tmp2f           ; <<8 x i8>> [#uses=1]
+  %tmp2gh = sub <8 x i8> %tmp2g, %tmp2h           ; <<8 x i8>> [#uses=1]
+  %tmp3gh = sub <8 x i8> zeroinitializer, %tmp3h  ; <<8 x i8>> [#uses=1]
+  %tmp4ef = sub <8 x i8> zeroinitializer, %tmp4g  ; <<8 x i8>> [#uses=1]
+  %tmp2efgh = mul <8 x i8> %tmp2ef, %tmp2gh       ; <<8 x i8>> [#uses=1]
+  %tmp3efgh = mul <8 x i8> undef, %tmp3gh         ; <<8 x i8>> [#uses=1]
+  %tmp4efgh = mul <8 x i8> %tmp4ef, undef         ; <<8 x i8>> [#uses=2]
+  call void @llvm.arm.neon.vst3.v8i8(i8* %A2, <8 x i8> %tmp4efgh, <8 x i8> %tmp3efgh, <8 x i8> %tmp2efgh)
+  %tmp4 = sub <8 x i8> %tmp4efgh, %tmp4abcd       ; <<8 x i8>> [#uses=1]
+  tail call void @llvm.arm.neon.vst3.v8i8(i8* %B, <8 x i8> zeroinitializer, <8 x i8> undef, <8 x i8> undef)
+  ret <8 x i8> %tmp4
+}