[llvm] r216248 - SROA: Handle a case of store size being smaller than allocation size

[Reid Kleckner]

Author: rnk
Date: Thu Aug 21 19:09:56 2014
New Revision: 216248

URL: http://llvm.org/viewvc/llvm-project?rev=216248&view=rev
Log:
SROA: Handle a case of store size being smaller than allocation size

In this case, we are creating an x86_fp80 slice for a union from C where
the padding bytes may contain real data. An x86_fp80 alloca is 16 bytes,
and that's just fine. We can't, however, use regular loads and stores to
access the slice, because the store size is only 10 bytes / 80 bits.
Instead, use memcpy and memset.

Fixes PR18726.

Reviewed By: chandlerc

Differential Revision: http://reviews.llvm.org/D5012

Modified:
    llvm/trunk/lib/Transforms/Scalar/SROA.cpp
    llvm/trunk/test/Transforms/SROA/slice-width.ll

Modified: llvm/trunk/lib/Transforms/Scalar/SROA.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/SROA.cpp?rev=216248&r1=216247&r2=216248&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/SROA.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/SROA.cpp Thu Aug 21 19:09:56 2014
@@ -2422,6 +2422,7 @@ private:
     if (!VecTy && !IntTy &&
         (BeginOffset > NewAllocaBeginOffset ||
          EndOffset < NewAllocaEndOffset ||
+         SliceSize != DL.getTypeStoreSize(AllocaTy) ||
          !AllocaTy->isSingleValueType() ||
          !DL.isLegalInteger(DL.getTypeSizeInBits(ScalarTy)) ||
          DL.getTypeSizeInBits(ScalarTy)%8 != 0)) {
@@ -2544,10 +2545,11 @@ private:
 
     // If this doesn't map cleanly onto the alloca type, and that type isn't
     // a single value type, just emit a memcpy.
-    bool EmitMemCpy
-      = !VecTy && !IntTy && (BeginOffset > NewAllocaBeginOffset ||
-                             EndOffset < NewAllocaEndOffset ||
-                             !NewAI.getAllocatedType()->isSingleValueType());
+    bool EmitMemCpy =
+        !VecTy && !IntTy &&
+        (BeginOffset > NewAllocaBeginOffset || EndOffset < NewAllocaEndOffset ||
+         SliceSize != DL.getTypeStoreSize(NewAI.getAllocatedType()) ||
+         !NewAI.getAllocatedType()->isSingleValueType());
 
     // If we're just going to emit a memcpy, the alloca hasn't changed, and the
     // size hasn't been shrunk based on analysis of the viable range, this is

Modified: llvm/trunk/test/Transforms/SROA/slice-width.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SROA/slice-width.ll?rev=216248&r1=216247&r2=216248&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/SROA/slice-width.ll (original)
+++ llvm/trunk/test/Transforms/SROA/slice-width.ll Thu Aug 21 19:09:56 2014
@@ -1,7 +1,8 @@
 ; RUN: opt < %s -sroa -S | FileCheck %s
-target datalayout = "e-p:64:64:64-p1:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64"
+target datalayout = "e-p:64:64:64-p1:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-f80:128-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64"
 
 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
+declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i32, i1) nounwind
 
 define void @no_split_on_non_byte_width(i32) {
 ; This tests that allocas are not split into slices that are not byte width multiple
@@ -23,3 +24,49 @@ load_i1:
   %t1 = load i1* %p1
   ret void
 }
+
+; PR18726: Check that we use memcpy and memset to fill out padding when we have
+; a slice with a simple single type whose store size is smaller than the slice
+; size.
+
+%union.Foo = type { x86_fp80, i64, i64 }
+
+ at foo_copy_source = external constant %union.Foo
+ at i64_sink = global i64 0
+
+define void @memcpy_fp80_padding() {
+  %x = alloca %union.Foo
+
+  ; Copy from a global.
+  %x_i8 = bitcast %union.Foo* %x to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %x_i8, i8* bitcast (%union.Foo* @foo_copy_source to i8*), i32 32, i32 16, i1 false)
+
+  ; Access a slice of the alloca to trigger SROA.
+  %mid_p = getelementptr %union.Foo* %x, i32 0, i32 1
+  %elt = load i64* %mid_p
+  store i64 %elt, i64* @i64_sink
+  ret void
+}
+; CHECK-LABEL: define void @memcpy_fp80_padding
+; CHECK: alloca x86_fp80
+; CHECK: call void @llvm.memcpy.p0i8.p0i8.i32
+; CHECK: load i64* getelementptr inbounds (%union.Foo* @foo_copy_source, i64 0, i32 1)
+; CHECK: load i64* getelementptr inbounds (%union.Foo* @foo_copy_source, i64 0, i32 2)
+
+define void @memset_fp80_padding() {
+  %x = alloca %union.Foo
+
+  ; Set to all ones.
+  %x_i8 = bitcast %union.Foo* %x to i8*
+  call void @llvm.memset.p0i8.i32(i8* %x_i8, i8 -1, i32 32, i32 16, i1 false)
+
+  ; Access a slice of the alloca to trigger SROA.
+  %mid_p = getelementptr %union.Foo* %x, i32 0, i32 1
+  %elt = load i64* %mid_p
+  store i64 %elt, i64* @i64_sink
+  ret void
+}
+; CHECK-LABEL: define void @memset_fp80_padding
+; CHECK: alloca x86_fp80
+; CHECK: call void @llvm.memset.p0i8.i32(i8* %{{.*}}, i8 -1, i32 16, i32 16, i1 false)
+; CHECK: store i64 -1, i64* @i64_sink