-; RUN: llc < %s -march=arm | FileCheck %s
-; RUN: llc < %s -march=thumb | FileCheck -check-prefix=THUMB %s
-; RUN: llc < %s -march=thumb -mattr=+thumb2 | FileCheck -check-prefix=T2 %s
-; XFAIL: *
+; RUN: llc -mtriple=arm-eabi -arm-atomic-cfg-tidy=0 %s -o - | FileCheck -check-prefix=ARM %s
+; RUN: llc -mtriple=thumb-eabi -arm-atomic-cfg-tidy=0 %s -o - | FileCheck -check-prefix=THUMB %s
+; RUN: llc -mtriple=thumb-eabi -arm-atomic-cfg-tidy=0 -mcpu=arm1156t2-s -mattr=+thumb2 %s -o - \
+; RUN: | FileCheck -check-prefix=T2 %s
+; RUN: llc -mtriple=thumbv8-eabi -arm-atomic-cfg-tidy=0 %s -o - | FileCheck -check-prefix=V8 %s
+
+; FIXME: The -march=thumb test doesn't change if -disable-peephole is specified.
%struct.Foo = type { i8* }
-define %struct.Foo* @_ZN3Foo7collectEj(%struct.Foo* %this, i32 %acc) nounwind readonly align 2 {
+; ARM-LABEL: foo:
+; THUMB-LABEL: foo:
+; T2-LABEL: foo:
+define %struct.Foo* @foo(%struct.Foo* %this, i32 %acc) nounwind readonly align 2 {
entry:
- %scevgep = getelementptr %struct.Foo* %this, i32 1
+ %scevgep = getelementptr %struct.Foo, %struct.Foo* %this, i32 1
br label %tailrecurse
tailrecurse: ; preds = %sw.bb, %entry
%lsr.iv = phi i32 [ %lsr.iv.next, %sw.bb ], [ 1, %entry ]
%acc.tr = phi i32 [ %or, %sw.bb ], [ %acc, %entry ]
%lsr.iv24 = bitcast %struct.Foo* %lsr.iv2 to i8**
- %scevgep5 = getelementptr i8** %lsr.iv24, i32 -1
- %tmp2 = load i8** %scevgep5
+ %scevgep5 = getelementptr i8*, i8** %lsr.iv24, i32 -1
+ %tmp2 = load i8*, i8** %scevgep5
%0 = ptrtoint i8* %tmp2 to i32
-; CHECK: ands r12, r12, #3
-; CHECK-NEXT: beq
+; ARM: ands {{r[0-9]+}}, {{r[0-9]+}}, #3
+; ARM-NEXT: beq
-; THUMB: movs r5, #3
-; THUMB-NEXT: mov r6, r4
-; THUMB-NEXT: ands r6, r5
-; THUMB-NEXT: tst r4, r5
+; THUMB: movs r[[R0:[0-9]+]], #3
+; THUMB-NEXT: ands r[[R0]], r
+; THUMB-NEXT: cmp r[[R0]], #0
; THUMB-NEXT: beq
-; T2: ands r12, r12, #3
+; T2: ands {{r[0-9]+}}, {{r[0-9]+}}, #3
; T2-NEXT: beq
%and = and i32 %0, 3
br i1 %tst, label %sw.bb, label %tailrecurse.switch
tailrecurse.switch: ; preds = %tailrecurse
+; V8-LABEL: %tailrecurse.switch
+; V8: cmp
+; V8-NEXT: beq
+; V8-NEXT: %tailrecurse.switch
+; V8: cmp
+; V8-NEXT: beq
+; V8-NEXT: %tailrecurse.switch
+; V8: cmp
+; V8-NEXT: bne
+; V8-NEXT: b
+; The trailing space in the last line checks that the branch is unconditional
switch i32 %and, label %sw.epilog [
i32 1, label %sw.bb
i32 3, label %sw.bb6
i32 2, label %sw.bb8
- ]
+ ], !prof !1
sw.bb: ; preds = %tailrecurse.switch, %tailrecurse
%shl = shl i32 %acc.tr, 1
%or = or i32 %and, %shl
%lsr.iv.next = add i32 %lsr.iv, 1
- %scevgep3 = getelementptr %struct.Foo* %lsr.iv2, i32 1
+ %scevgep3 = getelementptr %struct.Foo, %struct.Foo* %lsr.iv2, i32 1
br label %tailrecurse
sw.bb6: ; preds = %tailrecurse.switch
sw.bb8: ; preds = %tailrecurse.switch
%tmp1 = add i32 %acc.tr, %lsr.iv
- %add.ptr11 = getelementptr inbounds %struct.Foo* %this, i32 %tmp1
+ %add.ptr11 = getelementptr inbounds %struct.Foo, %struct.Foo* %this, i32 %tmp1
ret %struct.Foo* %add.ptr11
sw.epilog: ; preds = %tailrecurse.switch
%struct.S = type { i8* (i8*)*, [1 x i8] }
-; CHECK: bar
-; THUMB: bar
-; T2: bar
+; ARM-LABEL: bar:
+; THUMB-LABEL: bar:
+; T2-LABEL: bar:
+; V8-LABEL: bar:
define internal zeroext i8 @bar(%struct.S* %x, %struct.S* nocapture %y) nounwind readonly {
entry:
- %0 = getelementptr inbounds %struct.S* %x, i32 0, i32 1, i32 0
- %1 = load i8* %0, align 1
+ %0 = getelementptr inbounds %struct.S, %struct.S* %x, i32 0, i32 1, i32 0
+ %1 = load i8, i8* %0, align 1
%2 = zext i8 %1 to i32
-; CHECK: ands
+; ARM: ands
; THUMB: ands
; T2: ands
+; V8: ands
+; V8-NEXT: beq
%3 = and i32 %2, 112
%4 = icmp eq i32 %3, 0
br i1 %4, label %return, label %bb
bb: ; preds = %entry
- %5 = getelementptr inbounds %struct.S* %y, i32 0, i32 1, i32 0
- %6 = load i8* %5, align 1
+; V8-NEXT: %bb
+ %5 = getelementptr inbounds %struct.S, %struct.S* %y, i32 0, i32 1, i32 0
+ %6 = load i8, i8* %5, align 1
%7 = zext i8 %6 to i32
-; CHECK: andsne
+; ARM: andsne
; THUMB: ands
; T2: andsne
+; V8: ands
+; V8-NEXT: beq
%8 = and i32 %7, 112
%9 = icmp eq i32 %8, 0
br i1 %9, label %return, label %bb2
bb2: ; preds = %bb
+; V8-NEXT: %bb2
+; V8-NEXT: cmp
+; V8-NEXT: it ne
+; V8-NEXT: cmpne
+; V8-NEXT: bne
%10 = icmp eq i32 %3, 16
%11 = icmp eq i32 %8, 16
%or.cond = or i1 %10, %11
return: ; preds = %bb2, %bb, %entry
ret i8 1
}
+
+
+; We were looking through multiple COPY instructions to find an AND we might
+; fold into a TST, but in doing so we changed the register being tested allowing
+; folding of unrelated tests (in this case, a TST against r1 was eliminated in
+; favour of an AND of r0).
+
+; ARM-LABEL: test_tst_assessment:
+; THUMB-LABEL: test_tst_assessment:
+; T2-LABEL: test_tst_assessment:
+; V8-LABEL: test_tst_assessment:
+define i32 @test_tst_assessment(i1 %lhs, i1 %rhs) {
+ %lhs32 = zext i1 %lhs to i32
+ %rhs32 = zext i1 %rhs to i32
+ %diff = sub nsw i32 %lhs32, %rhs32
+; ARM: tst r1, #1
+; THUMB: movs [[RTMP:r[0-9]+]], #1
+; THUMB: tst r1, [[RTMP]]
+; T2: tst.w r1, #1
+; V8: tst.w r1, #1
+ ret i32 %diff
+}
+
+!1 = !{!"branch_weights", i32 1, i32 1, i32 3, i32 2 }