R600: Fix calculation of stack offset in AMDGPUFrameLowering

We weren't computing structure size correctly and we were relying on
the original alloca instruction to compute the offset, which isn't
always reliable.

Reviewed-by: Vincent Lejeune <vljn@ovi.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183568 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/AMDGPUFrameLowering.cpp b/lib/Target/R600/AMDGPUFrameLowering.cpp
index 815d6f71c3be69749d11c8f265c2a7ace2505a11..40f14d2f67c51626e29872ea711a1b4b0de9cfa7 100644 (file)
--- a/lib/Target/R600/AMDGPUFrameLowering.cpp
+++ b/lib/Target/R600/AMDGPUFrameLowering.cpp
@@ -78,27 +78,8 @@ int AMDGPUFrameLowering::getFrameIndexOffset(const MachineFunction &MF,
   int UpperBound = FI == -1 ? MFI->getNumObjects() : FI;
 
   for (int i = MFI->getObjectIndexBegin(); i < UpperBound; ++i) {
-    const AllocaInst *Alloca = MFI->getObjectAllocation(i);
-    unsigned ArrayElements;
-    const Type *AllocaType = Alloca->getAllocatedType();
-    const Type *ElementType;
-
-    if (AllocaType->isArrayTy()) {
-      ArrayElements = AllocaType->getArrayNumElements();
-      ElementType = AllocaType->getArrayElementType();
-    } else {
-      ArrayElements = 1;
-      ElementType = AllocaType;
-    }
-
-    unsigned VectorElements;
-    if (ElementType->isVectorTy()) {
-      VectorElements = ElementType->getVectorNumElements();
-    } else {
-      VectorElements = 1;
-    }
-
-    Offset += (VectorElements / getStackWidth(MF)) * ArrayElements;
+    unsigned Size = MFI->getObjectSize(i);
+    Offset += (Size / (getStackWidth(MF) * 4));
   }
   return Offset;
 }

diff --git a/test/CodeGen/R600/indirect-addressing.ll b/test/CodeGen/R600/indirect-addressing.ll
index 7291cb42e7fb9d0982309accc6e833d963e5a5fe..bd72cd96e2e5c2fada0e445c00de6d8173fda167 100644 (file)
--- a/test/CodeGen/R600/indirect-addressing.ll
+++ b/test/CodeGen/R600/indirect-addressing.ll
@@ -30,3 +30,36 @@ entry:
   store i32 %3, i32 addrspace(1)* %arrayidx13
   ret void
 }
+
+; This test checks that the stack offset is calculated correctly for structs.
+; All register loads/stores should be optimized away, so there shouldn't be
+; any MOVA instructions.
+;
+; XXX: This generated code has unnecessary MOVs, we should be able to optimize
+; this.
+
+; CHECK: @multiple_structs
+; CHECK-NOT: MOVA_INT
+
+%struct.point = type { i32, i32 }
+
+define void @multiple_structs(i32 addrspace(1)* %out) {
+entry:
+  %a = alloca %struct.point
+  %b = alloca %struct.point
+  %a.x.ptr = getelementptr %struct.point* %a, i32 0, i32 0
+  %a.y.ptr = getelementptr %struct.point* %a, i32 0, i32 1
+  %b.x.ptr = getelementptr %struct.point* %b, i32 0, i32 0
+  %b.y.ptr = getelementptr %struct.point* %b, i32 0, i32 1
+  store i32 0, i32* %a.x.ptr
+  store i32 1, i32* %a.y.ptr
+  store i32 2, i32* %b.x.ptr
+  store i32 3, i32* %b.y.ptr
+  %a.indirect.ptr = getelementptr %struct.point* %a, i32 0, i32 0
+  %b.indirect.ptr = getelementptr %struct.point* %b, i32 0, i32 0
+  %a.indirect = load i32* %a.indirect.ptr
+  %b.indirect = load i32* %b.indirect.ptr
+  %0 = add i32 %a.indirect, %b.indirect
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}