Enable tail calls on ARM by default, with some

basic tests.

This has been well tested on Darwin but not elsewhere.
It should work provided the linker correctly resolves
  B.W  <label in other function>
which it has not seen before, at least from llvm-based
compilers.  I'm leaving the arm-tail-calls switch in
until I see if there's any problems because of that;
it might need to be disabled for some environments.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106299 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 221 insertions, 1 deletions

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 50258c6..216ba65 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -55,7 +55,7 @@ STATISTIC(NumTailCalls, "Number of tail calls");
 static cl::opt<bool>
 EnableARMTailCalls("arm-tail-calls", cl::Hidden,
   cl::desc("Generate tail calls (TEMPORARY OPTION)."),
-  cl::init(false));
+  cl::init(true));
 
 static cl::opt<bool>
 EnableARMLongCalls("arm-long-calls", cl::Hidden,
diff --git a/test/CodeGen/ARM/call-tc.ll b/test/CodeGen/ARM/call-tc.ll
new file mode 100644
index 0000000..8103fab
--- /dev/null
+++ b/test/CodeGen/ARM/call-tc.ll
@@ -0,0 +1,36 @@
+; RUN: llc < %s -march=arm | FileCheck %s -check-prefix=CHECKV4
+; RUN: llc < %s -march=arm -mattr=+v5t | FileCheck %s -check-prefix=CHECKV5
+; RUN: llc < %s -march=arm -mtriple=arm-linux-gnueabi\
+; RUN:   -relocation-model=pic | FileCheck %s -check-prefix=CHECKELF
+
+@t = weak global i32 ()* null           ; <i32 ()**> [#uses=1]
+
+declare void @g(i32, i32, i32, i32)
+
+define void @f() {
+; CHECKELF: PLT
+        call void @g( i32 1, i32 2, i32 3, i32 4 )
+        ret void
+}
+
+define void @g.upgrd.1() {
+; CHECKV4: bx r0 @ TAILCALL
+; CHECKV5: bx r0 @ TAILCALL
+        %tmp = load i32 ()** @t         ; <i32 ()*> [#uses=1]
+        %tmp.upgrd.2 = tail call i32 %tmp( )            ; <i32> [#uses=0]
+        ret void
+}
+
+define i32* @m_231b(i32, i32, i32*, i32*, i32*) nounwind {
+; CHECKV4: m_231b
+; CHECKV4: bx r{{.*}}
+BB0:
+  %5 = inttoptr i32 %0 to i32*                    ; <i32*> [#uses=1]
+  %t35 = volatile load i32* %5                    ; <i32> [#uses=1]
+  %6 = inttoptr i32 %t35 to i32**                 ; <i32**> [#uses=1]
+  %7 = getelementptr i32** %6, i32 86             ; <i32**> [#uses=1]
+  %8 = load i32** %7                              ; <i32*> [#uses=1]
+  %9 = bitcast i32* %8 to i32* (i32, i32*, i32, i32*, i32*, i32*)* ; <i32* (i32, i32*, i32, i32*, i32*, i32*)*> [#uses=1]
+  %10 = call i32* %9(i32 %0, i32* null, i32 %1, i32* %2, i32* %3, i32* %4) ; <i32*> [#uses=1]
+  ret i32* %10
+}
diff --git a/test/CodeGen/ARM/ifcvt6-tc.ll b/test/CodeGen/ARM/ifcvt6-tc.ll
new file mode 100644
index 0000000..5b28804
--- /dev/null
+++ b/test/CodeGen/ARM/ifcvt6-tc.ll
@@ -0,0 +1,23 @@
+; RUN: llc < %s -march=arm -mtriple=arm-apple-darwin | \
+; RUN:   grep cmpne | count 1
+; RUN: llc < %s -march=arm -mtriple=arm-apple-darwin | \
+; RUN:   grep bhi | count 1
+; Here, tail call wins over eliminating branches.  It is 1 fewer instruction
+; and removes all stack accesses, so seems like a win.
+
+define void @foo(i32 %X, i32 %Y) {
+entry:
+	%tmp1 = icmp ult i32 %X, 4		; <i1> [#uses=1]
+	%tmp4 = icmp eq i32 %Y, 0		; <i1> [#uses=1]
+	%tmp7 = or i1 %tmp4, %tmp1		; <i1> [#uses=1]
+	br i1 %tmp7, label %cond_true, label %UnifiedReturnBlock
+
+cond_true:		; preds = %entry
+	%tmp10 = tail call i32 (...)* @bar( )		; <i32> [#uses=0]
+	ret void
+
+UnifiedReturnBlock:		; preds = %entry
+	ret void
+}
+
+declare i32 @bar(...)
diff --git a/test/CodeGen/ARM/insn-sched1-tc.ll b/test/CodeGen/ARM/insn-sched1-tc.ll
new file mode 100644
index 0000000..c457c8c
--- /dev/null
+++ b/test/CodeGen/ARM/insn-sched1-tc.ll
@@ -0,0 +1,11 @@
+; RUN: llc < %s -march=arm -mattr=+v6
+; RUN: llc < %s -mtriple=arm-apple-darwin -mattr=+v6 |\
+; RUN:   grep mov | count 2
+
+define i32 @test(i32 %x) {
+        %tmp = trunc i32 %x to i16              ; <i16> [#uses=1]
+        %tmp2 = tail call i32 @f( i32 1, i16 %tmp )             ; <i32> [#uses=1]
+        ret i32 %tmp2
+}
+
+declare i32 @f(i32, i16)
diff --git a/test/CodeGen/ARM/ldm-tc.ll b/test/CodeGen/ARM/ldm-tc.ll
new file mode 100644
index 0000000..3819192
--- /dev/null
+++ b/test/CodeGen/ARM/ldm-tc.ll
@@ -0,0 +1,37 @@
+; RUN: llc < %s -mtriple=arm-apple-darwin | FileCheck %s
+
+@X = external global [0 x i32]          ; <[0 x i32]*> [#uses=5]
+
+define i32 @t1() {
+; CHECK: t1:
+; CHECK: ldmia
+        %tmp = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 0)            ; <i32> [#uses=1]
+        %tmp3 = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 1)           ; <i32> [#uses=1]
+        %tmp4 = tail call i32 @f1( i32 %tmp, i32 %tmp3 )                ; <i32> [#uses=1]
+        ret i32 %tmp4
+}
+
+define i32 @t2() {
+; CHECK: t2:
+; CHECK: ldmia
+        %tmp = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 2)            ; <i32> [#uses=1]
+        %tmp3 = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 3)           ; <i32> [#uses=1]
+        %tmp5 = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 4)           ; <i32> [#uses=1]
+        %tmp6 = tail call i32 @f2( i32 %tmp, i32 %tmp3, i32 %tmp5 )             ; <i32> [#uses=1]
+        ret i32 %tmp6
+}
+
+define i32 @t3() {
+; CHECK: t3:
+; CHECK: ldmib
+; CHECK: b.w _f2 @ TAILCALL
+        %tmp = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 1)            ; <i32> [#uses=1]
+        %tmp3 = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 2)           ; <i32> [#uses=1]
+        %tmp5 = load i32* getelementptr ([0 x i32]* @X, i32 0, i32 3)           ; <i32> [#uses=1]
+        %tmp6 = tail call i32 @f2( i32 %tmp, i32 %tmp3, i32 %tmp5 )             ; <i32> [#uses=1]
+        ret i32 %tmp6
+}
+
+declare i32 @f1(i32, i32)
+
+declare i32 @f2(i32, i32, i32)
diff --git a/test/CodeGen/Thumb2/thumb2-call-tc.ll b/test/CodeGen/Thumb2/thumb2-call-tc.ll
new file mode 100644
index 0000000..d31ae0c
--- /dev/null
+++ b/test/CodeGen/Thumb2/thumb2-call-tc.ll
@@ -0,0 +1,27 @@
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mattr=+thumb2 | FileCheck %s -check-prefix=DARWIN
+; RUN: llc < %s -mtriple=thumbv7-linux -mattr=+thumb2 | FileCheck %s -check-prefix=LINUX
+
+@t = weak global i32 ()* null           ; <i32 ()**> [#uses=1]
+
+declare void @g(i32, i32, i32, i32)
+
+define void @f() {
+; DARWIN: f:
+; DARWIN: blx _g
+
+; LINUX: f:
+; LINUX: bl g
+        call void @g( i32 1, i32 2, i32 3, i32 4 )
+        ret void
+}
+
+define void @h() {
+; DARWIN: h:
+; DARWIN: bx r0 @ TAILCALL
+
+; LINUX: h:
+; LINUX: bx r0 @ TAILCALL
+        %tmp = load i32 ()** @t         ; <i32 ()*> [#uses=1]
+        %tmp.upgrd.2 = tail call i32 %tmp( )            ; <i32> [#uses=0]
+        ret void
+}
diff --git a/test/CodeGen/Thumb2/thumb2-ifcvt1-tc.ll b/test/CodeGen/Thumb2/thumb2-ifcvt1-tc.ll
new file mode 100644
index 0000000..c024415
--- /dev/null
+++ b/test/CodeGen/Thumb2/thumb2-ifcvt1-tc.ll
@@ -0,0 +1,86 @@
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin | FileCheck %s
+
+define i32 @t1(i32 %a, i32 %b, i32 %c, i32 %d) nounwind {
+; CHECK: t1:
+; CHECK: it ne
+; CHECK: cmpne
+	switch i32 %c, label %cond_next [
+		 i32 1, label %cond_true
+		 i32 7, label %cond_true
+	]
+
+cond_true:
+	%tmp12 = add i32 %a, 1
+	%tmp1518 = add i32 %tmp12, %b
+	ret i32 %tmp1518
+
+cond_next:
+	%tmp15 = add i32 %b, %a
+	ret i32 %tmp15
+}
+
+; FIXME: Check for # of unconditional branch after adding branch folding post ifcvt.
+define i32 @t2(i32 %a, i32 %b) nounwind {
+entry:
+; CHECK: t2:
+; CHECK: ite gt
+; CHECK: subgt
+; CHECK: suble
+	%tmp1434 = icmp eq i32 %a, %b		; <i1> [#uses=1]
+	br i1 %tmp1434, label %bb17, label %bb.outer
+
+bb.outer:		; preds = %cond_false, %entry
+	%b_addr.021.0.ph = phi i32 [ %b, %entry ], [ %tmp10, %cond_false ]		; <i32> [#uses=5]
+	%a_addr.026.0.ph = phi i32 [ %a, %entry ], [ %a_addr.026.0, %cond_false ]		; <i32> [#uses=1]
+	br label %bb
+
+bb:		; preds = %cond_true, %bb.outer
+	%indvar = phi i32 [ 0, %bb.outer ], [ %indvar.next, %cond_true ]		; <i32> [#uses=2]
+	%tmp. = sub i32 0, %b_addr.021.0.ph		; <i32> [#uses=1]
+	%tmp.40 = mul i32 %indvar, %tmp.		; <i32> [#uses=1]
+	%a_addr.026.0 = add i32 %tmp.40, %a_addr.026.0.ph		; <i32> [#uses=6]
+	%tmp3 = icmp sgt i32 %a_addr.026.0, %b_addr.021.0.ph		; <i1> [#uses=1]
+	br i1 %tmp3, label %cond_true, label %cond_false
+
+cond_true:		; preds = %bb
+	%tmp7 = sub i32 %a_addr.026.0, %b_addr.021.0.ph		; <i32> [#uses=2]
+	%tmp1437 = icmp eq i32 %tmp7, %b_addr.021.0.ph		; <i1> [#uses=1]
+	%indvar.next = add i32 %indvar, 1		; <i32> [#uses=1]
+	br i1 %tmp1437, label %bb17, label %bb
+
+cond_false:		; preds = %bb
+	%tmp10 = sub i32 %b_addr.021.0.ph, %a_addr.026.0		; <i32> [#uses=2]
+	%tmp14 = icmp eq i32 %a_addr.026.0, %tmp10		; <i1> [#uses=1]
+	br i1 %tmp14, label %bb17, label %bb.outer
+
+bb17:		; preds = %cond_false, %cond_true, %entry
+	%a_addr.026.1 = phi i32 [ %a, %entry ], [ %tmp7, %cond_true ], [ %a_addr.026.0, %cond_false ]		; <i32> [#uses=1]
+	ret i32 %a_addr.026.1
+}
+
+@x = external global i32*		; <i32**> [#uses=1]
+
+define void @foo(i32 %a) nounwind {
+entry:
+	%tmp = load i32** @x		; <i32*> [#uses=1]
+	store i32 %a, i32* %tmp
+	ret void
+}
+
+; Tail call prevents use of ifcvt in this one.  Seems like a win though.
+define void @t3(i32 %a, i32 %b) nounwind {
+entry:
+; CHECK: t3:
+; CHECK-NOT: it lt
+; CHECK-NOT: poplt
+; CHECK: b.w _foo @ TAILCALL
+	%tmp1 = icmp sgt i32 %a, 10		; <i1> [#uses=1]
+	br i1 %tmp1, label %cond_true, label %UnifiedReturnBlock
+
+cond_true:		; preds = %entry
+	tail call void @foo( i32 %b )
+	ret void
+
+UnifiedReturnBlock:		; preds = %entry
+	ret void
+}