[Hexagon] Implement frame pointer elimination with -fomit-frame-pointer

It applies to leaf functions that are otherwise not required to have
a frame pointer.

llvm-svn: 306888

3 files changed, 92 insertions, 24 deletions

diff --git a/llvm/test/CodeGen/Hexagon/fpelim-basic.ll b/llvm/test/CodeGen/Hexagon/fpelim-basic.ll
new file mode 100644
index 00000000000..ffec07f7dbf
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/fpelim-basic.ll
@@ -0,0 +1,91 @@
+; RUN: llc -march=hexagon < %s | FileCheck %s
+
+target triple = "hexagon"
+
+; FP elimination enabled.
+;
+; CHECK-LABEL: danny:
+; CHECK: r29 = add(r29,#-[[SIZE:[0-9]+]])
+; CHECK: r29 = add(r29,#[[SIZE]])
+define i32 @danny(i32 %a0, i32 %a1) local_unnamed_addr #0 {
+b2:
+  %v3 = alloca [32 x i32], align 8
+  %v4 = bitcast [32 x i32]* %v3 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 128, i8* nonnull %v4) #3
+  br label %b5
+
+b5:                                               ; preds = %b5, %b2
+  %v6 = phi i32 [ 0, %b2 ], [ %v8, %b5 ]
+  %v7 = getelementptr inbounds [32 x i32], [32 x i32]* %v3, i32 0, i32 %v6
+  store i32 %v6, i32* %v7, align 4
+  %v8 = add nuw nsw i32 %v6, 1
+  %v9 = icmp eq i32 %v8, 32
+  br i1 %v9, label %b10, label %b5
+
+b10:                                              ; preds = %b5
+  %v11 = getelementptr inbounds [32 x i32], [32 x i32]* %v3, i32 0, i32 %a0
+  store i32 %a1, i32* %v11, align 4
+  br label %b12
+
+b12:                                              ; preds = %b12, %b10
+  %v13 = phi i32 [ 0, %b10 ], [ %v18, %b12 ]
+  %v14 = phi i32 [ 0, %b10 ], [ %v17, %b12 ]
+  %v15 = getelementptr inbounds [32 x i32], [32 x i32]* %v3, i32 0, i32 %v13
+  %v16 = load i32, i32* %v15, align 4
+  %v17 = add nsw i32 %v16, %v14
+  %v18 = add nuw nsw i32 %v13, 1
+  %v19 = icmp eq i32 %v18, 32
+  br i1 %v19, label %b20, label %b12
+
+b20:                                              ; preds = %b12
+  call void @llvm.lifetime.end.p0i8(i64 128, i8* nonnull %v4) #3
+  ret i32 %v17
+}
+
+; FP elimination disabled.
+;
+; CHECK-LABEL: sammy:
+; CHECK: allocframe
+; CHECK: dealloc_return
+define i32 @sammy(i32 %a0, i32 %a1) local_unnamed_addr #1 {
+b2:
+  %v3 = alloca [32 x i32], align 8
+  %v4 = bitcast [32 x i32]* %v3 to i8*
+  call void @llvm.lifetime.start.p0i8(i64 128, i8* nonnull %v4) #3
+  br label %b5
+
+b5:                                               ; preds = %b5, %b2
+  %v6 = phi i32 [ 0, %b2 ], [ %v8, %b5 ]
+  %v7 = getelementptr inbounds [32 x i32], [32 x i32]* %v3, i32 0, i32 %v6
+  store i32 %v6, i32* %v7, align 4
+  %v8 = add nuw nsw i32 %v6, 1
+  %v9 = icmp eq i32 %v8, 32
+  br i1 %v9, label %b10, label %b5
+
+b10:                                              ; preds = %b5
+  %v11 = getelementptr inbounds [32 x i32], [32 x i32]* %v3, i32 0, i32 %a0
+  store i32 %a1, i32* %v11, align 4
+  br label %b12
+
+b12:                                              ; preds = %b12, %b10
+  %v13 = phi i32 [ 0, %b10 ], [ %v18, %b12 ]
+  %v14 = phi i32 [ 0, %b10 ], [ %v17, %b12 ]
+  %v15 = getelementptr inbounds [32 x i32], [32 x i32]* %v3, i32 0, i32 %v13
+  %v16 = load i32, i32* %v15, align 4
+  %v17 = add nsw i32 %v16, %v14
+  %v18 = add nuw nsw i32 %v13, 1
+  %v19 = icmp eq i32 %v18, 32
+  br i1 %v19, label %b20, label %b12
+
+b20:                                              ; preds = %b12
+  call void @llvm.lifetime.end.p0i8(i64 128, i8* nonnull %v4) #3
+  ret i32 %v17
+}
+
+declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #2
+declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #2
+
+attributes #0 = { nounwind readnone "no-frame-pointer-elim"="false" "target-cpu"="hexagonv60" }
+attributes #1 = { nounwind readnone "no-frame-pointer-elim"="true" "target-cpu"="hexagonv60" }
+attributes #2 = { argmemonly nounwind }
+attributes #3 = { nounwind }
diff --git a/llvm/test/CodeGen/Hexagon/frame.ll b/llvm/test/CodeGen/Hexagon/frame.ll
deleted file mode 100644
index e87acb8cd79..00000000000
--- a/llvm/test/CodeGen/Hexagon/frame.ll
+++ /dev/null
@@ -1,23 +0,0 @@
-; RUN: llc -march=hexagon -mcpu=hexagonv4 < %s | FileCheck %s
-
-@num = external global i32
-@acc = external global i32
-@num2 = external global i32
-
-; CHECK: allocframe
-; CHECK: dealloc_return
-
-define i32 @foo() nounwind {
-entry:
-  %i = alloca i32, align 4
-  %0 = load i32, i32* @num, align 4
-  store i32 %0, i32* %i, align 4
-  %1 = load i32, i32* %i, align 4
-  %2 = load i32, i32* @acc, align 4
-  %mul = mul nsw i32 %1, %2
-  %3 = load i32, i32* @num2, align 4
-  %add = add nsw i32 %mul, %3
-  store i32 %add, i32* %i, align 4
-  %4 = load i32, i32* %i, align 4
-  ret i32 %4
-}
diff --git a/llvm/test/CodeGen/Hexagon/newvaluejump2.ll b/llvm/test/CodeGen/Hexagon/newvaluejump2.ll
index 4c897f0830f..fbc3f2925d1 100644
--- a/llvm/test/CodeGen/Hexagon/newvaluejump2.ll
+++ b/llvm/test/CodeGen/Hexagon/newvaluejump2.ll
@@ -6,7 +6,7 @@
 @Reg = common global i32 0, align 4
 define i32 @main() nounwind {
 entry:
-; CHECK: if (cmp.gt(r{{[0-9]+}},r{{[0-9]+}}.new)) jump:{{[t|nt]}} .LBB{{[0-9]+}}_{{[0-9]+}}
+; CHECK: if (cmp.gt(r{{[0-9]+}}.new,r{{[0-9]+}})) jump:{{[t|nt]}} .LBB{{[0-9]+}}_{{[0-9]+}}
   %Reg2 = alloca i32, align 4
   %0 = load i32, i32* %Reg2, align 4
   %1 = load i32, i32* @Reg, align 4