Fix a miscompile in the DAG combiner. Previously, we would incorrectly

try to reduce the width of this load, and would end up transforming:

  (truncate (lshr (sextload i48 <ptr> as i64), 32) to i32)
to
  (truncate (zextload i32 <ptr+4> as i64) to i32)

We lost the sext attached to the load while building the narrower i32
load, and replaced it with a zext because lshr always zext's the
results. Instead, bail out of this combine when there is a conflict
between a sextload and a zext narrowing. The rest of the DAG combiner
still optimize the code down to the proper single instruction:

  movswl 6(...),%eax

Which is exactly what we wanted. Previously we read past the end *and*
missed the sign extension:

  movl 6(...), %eax

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169802 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 23 insertions, 2 deletions

diff --git a/test/CodeGen/X86/sext-load.ll b/test/CodeGen/X86/sext-load.ll
index c9b39d3a48..58c93229a2 100644
--- a/test/CodeGen/X86/sext-load.ll
+++ b/test/CodeGen/X86/sext-load.ll
@@ -1,9 +1,30 @@
-; RUN: llc < %s -march=x86 | grep movsbl
+; RUN: llc < %s -march=x86 | FileCheck %s
 
-define i32 @foo(i32 %X) nounwind  {
+; When doing sign extension, use the sext-load lowering to take advantage of
+; x86's sign extension during loads.
+;
+; CHECK: test1:
+; CHECK:      movsbl {{.*}}, %eax
+; CHECK-NEXT: ret
+define i32 @test1(i32 %X) nounwind  {
 entry:
 	%tmp12 = trunc i32 %X to i8		; <i8> [#uses=1]
 	%tmp123 = sext i8 %tmp12 to i32		; <i32> [#uses=1]
 	ret i32 %tmp123
 }
 
+; When using a sextload representation, ensure that the sign extension is
+; preserved even when removing shifted-out low bits.
+;
+; CHECK: test2:
+; CHECK:      movswl {{.*}}, %eax
+; CHECK-NEXT: ret
+define i32 @test2({i16, [6 x i8]}* %this) {
+entry:
+  %b48 = getelementptr inbounds { i16, [6 x i8] }* %this, i32 0, i32 1
+  %cast = bitcast [6 x i8]* %b48 to i48*
+  %bf.load = load i48* %cast, align 2
+  %bf.ashr = ashr i48 %bf.load, 32
+  %bf.cast = trunc i48 %bf.ashr to i32
+  ret i32 %bf.cast
+}