Merging r323355:

------------------------------------------------------------------------
r323355 | nha | 2018-01-24 19:02:05 +0100 (Wed, 24 Jan 2018) | 9 lines

Revert r321751, "StructurizeCFG: Fix broken backedge detection"

It causes regressions in various OpenGL test suites.

Keep the test cases introduced by r321751 as XFAIL, and add a test case
for the regression.

Change-Id: I90b4cc354f68cebe5fcef1f2422dc8fe1c6d3514
Bugzilla: https://bugs.llvm.org/show_bug.cgi?id=36015
------------------------------------------------------------------------


git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_60@323749 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 123 insertions, 144 deletions

diff --git a/test/CodeGen/AMDGPU/multilevel-break.ll b/test/CodeGen/AMDGPU/multilevel-break.ll
index 5b556f12f0d..8cc02d49709 100644
--- a/test/CodeGen/AMDGPU/multilevel-break.ll
+++ b/test/CodeGen/AMDGPU/multilevel-break.ll
@@ -66,10 +66,9 @@ ENDIF:                                            ; preds = %LOOP
 
 ; OPT-LABEL: define amdgpu_kernel void @multi_if_break_loop(
 ; OPT: llvm.amdgcn.break
-; OPT: llvm.amdgcn.break
+; OPT: llvm.amdgcn.loop
 ; OPT: llvm.amdgcn.if.break
 ; OPT: llvm.amdgcn.if.break
-; OPT: llvm.amdgcn.loop
 ; OPT: llvm.amdgcn.end.cf
 
 ; GCN-LABEL: {{^}}multi_if_break_loop:
diff --git a/test/CodeGen/AMDGPU/nested-loop-conditions.ll b/test/CodeGen/AMDGPU/nested-loop-conditions.ll
index 96d2841e685..672549c8ea6 100644
--- a/test/CodeGen/AMDGPU/nested-loop-conditions.ll
+++ b/test/CodeGen/AMDGPU/nested-loop-conditions.ll
@@ -124,100 +124,55 @@ bb23:                                             ; preds = %bb10
 ; Earlier version of above, before a run of the structurizer.
 ; IR-LABEL: @nested_loop_conditions(
 
-; IR: %tmp1235 = icmp slt i32 %tmp1134, 9
-; IR:   br i1 %tmp1235, label %bb14.lr.ph, label %Flow
-
-; IR: bb14.lr.ph:
-; IR: br label %bb14
-
-; IR: Flow3:
-; IR:   call void @llvm.amdgcn.end.cf(i64 %18)
-; IR:   %0 = call { i1, i64 } @llvm.amdgcn.if(i1 %17)
-; IR:   %1 = extractvalue { i1, i64 } %0, 0
-; IR:   %2 = extractvalue { i1, i64 } %0, 1
-; IR:   br i1 %1, label %bb4.bb13_crit_edge, label %Flow4
-
-; IR: bb4.bb13_crit_edge:
-; IR:   br label %Flow4
-
-; IR: Flow4:
-; IR:   %3 = phi i1 [ true, %bb4.bb13_crit_edge ], [ false, %Flow3 ]
-; IR:   call void @llvm.amdgcn.end.cf(i64 %2)
-; IR:   br label %Flow
-
-; IR: bb13:
-; IR:   br label %bb31
-
-; IR: Flow:
-; IR:   %4 = phi i1 [ %3, %Flow4 ], [ true, %bb ]
-; IR:   %5 = call { i1, i64 } @llvm.amdgcn.if(i1 %4)
-; IR:   %6 = extractvalue { i1, i64 } %5, 0
-; IR:   %7 = extractvalue { i1, i64 } %5, 1
-; IR:   br i1 %6, label %bb13, label %bb31
-
-; IR: bb14:
-; IR:   %phi.broken = phi i64 [ %18, %Flow2 ], [ 0, %bb14.lr.ph ]
-; IR:   %tmp1037 = phi i32 [ %tmp1033, %bb14.lr.ph ], [ %16, %Flow2 ]
-; IR:   %tmp936 = phi <4 x i32> [ %tmp932, %bb14.lr.ph ], [ %15, %Flow2 ]
-; IR:   %tmp15 = icmp eq i32 %tmp1037, 1
-; IR:   %8 = xor i1 %tmp15, true
-; IR:   %9 = call { i1, i64 } @llvm.amdgcn.if(i1 %8)
-; IR:   %10 = extractvalue { i1, i64 } %9, 0
-; IR:   %11 = extractvalue { i1, i64 } %9, 1
-; IR:   br i1 %10, label %bb31.loopexit, label %Flow1
+; IR: Flow7:
+; IR-NEXT: call void @llvm.amdgcn.end.cf(i64 %17)
+; IR-NEXT: %0 = call { i1, i64 } @llvm.amdgcn.if(i1 %15)
+; IR-NEXT: %1 = extractvalue { i1, i64 } %0, 0
+; IR-NEXT: %2 = extractvalue { i1, i64 } %0, 1
+; IR-NEXT: br i1 %1, label %bb4.bb13_crit_edge, label %Flow8
 
 ; IR: Flow1:
-; IR:   %12 = call { i1, i64 } @llvm.amdgcn.else(i64 %11)
-; IR:   %13 = extractvalue { i1, i64 } %12, 0
-; IR:   %14 = extractvalue { i1, i64 } %12, 1
-; IR:   br i1 %13, label %bb16, label %Flow2
-
-; IR: bb16:
-; IR:   %tmp17 = bitcast i64 %tmp3 to <2 x i32>
-; IR:   br label %bb18
+; IR-NEXT: %loop.phi = phi i64 [ %loop.phi9, %Flow6 ], [ %phi.broken, %bb14 ]
+; IR-NEXT: %13 = phi <4 x i32> [ %29, %Flow6 ], [ undef, %bb14 ]
+; IR-NEXT: %14 = phi i32 [ %30, %Flow6 ], [ undef, %bb14 ]
+; IR-NEXT: %15 = phi i1 [ %31, %Flow6 ], [ false, %bb14 ]
+; IR-NEXT: %16 = phi i1 [ false, %Flow6 ], [ %8, %bb14 ]
+; IR-NEXT: %17 = call i64 @llvm.amdgcn.else.break(i64 %11, i64 %loop.phi)
+; IR-NEXT: call void @llvm.amdgcn.end.cf(i64 %11)
+; IR-NEXT: %18 = call i1 @llvm.amdgcn.loop(i64 %17)
+; IR-NEXT: br i1 %18, label %Flow7, label %bb14
 
 ; IR: Flow2:
-; IR:   %loop.phi = phi i64 [ %21, %bb21 ], [ %phi.broken, %Flow1 ]
-; IR:   %15 = phi <4 x i32> [ %tmp9, %bb21 ], [ undef, %Flow1 ]
-; IR:   %16 = phi i32 [ %tmp10, %bb21 ], [ undef, %Flow1 ]
-; IR:   %17 = phi i1 [ %20, %bb21 ], [ false, %Flow1 ]
-; IR:   %18 = call i64 @llvm.amdgcn.else.break(i64 %14, i64 %loop.phi)
-; IR:   call void @llvm.amdgcn.end.cf(i64 %14)
-; IR:   %19 = call i1 @llvm.amdgcn.loop(i64 %18)
-; IR:   br i1 %19, label %Flow3, label %bb14
-
-; IR: bb18:
-; IR:   %tmp19 = load volatile i32, i32 addrspace(1)* undef
-; IR:   %tmp20 = icmp slt i32 %tmp19, 9
-; IR:   br i1 %tmp20, label %bb21, label %bb18
+; IR-NEXT: %loop.phi10 = phi i64 [ %loop.phi11, %Flow5 ], [ %12, %bb16 ]
+; IR-NEXT: %19 = phi <4 x i32> [ %29, %Flow5 ], [ undef, %bb16 ]
+; IR-NEXT: %20 = phi i32 [ %30, %Flow5 ], [ undef, %bb16 ]
+; IR-NEXT: %21 = phi i1 [ %31, %Flow5 ], [ false, %bb16 ]
+; IR-NEXT: %22 = phi i1 [ false, %Flow5 ], [ false, %bb16 ]
+; IR-NEXT: %23 = phi i1 [ false, %Flow5 ], [ %8, %bb16 ]
+; IR-NEXT: %24 = call { i1, i64 } @llvm.amdgcn.if(i1 %23)
+; IR-NEXT: %25 = extractvalue { i1, i64 } %24, 0
+; IR-NEXT: %26 = extractvalue { i1, i64 } %24, 1
+; IR-NEXT: br i1 %25, label %bb21, label %Flow3
 
 ; IR: bb21:
-; IR:   %tmp22 = extractelement <2 x i32> %tmp17, i64 1
-; IR:   %tmp23 = lshr i32 %tmp22, 16
-; IR:   %tmp24 = select i1 undef, i32 undef, i32 %tmp23
-; IR:   %tmp25 = uitofp i32 %tmp24 to float
-; IR:   %tmp26 = fmul float %tmp25, 0x3EF0001000000000
-; IR:   %tmp27 = fsub float %tmp26, undef
-; IR:   %tmp28 = fcmp olt float %tmp27, 5.000000e-01
-; IR:   %tmp29 = select i1 %tmp28, i64 1, i64 2
-; IR:   %tmp30 = extractelement <4 x i32> %tmp936, i64 %tmp29
-; IR:   %tmp7 = zext i32 %tmp30 to i64
-; IR:   %tmp8 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* undef, i64 %tmp7
-; IR:   %tmp9 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp8, align 16
-; IR:   %tmp10 = extractelement <4 x i32> %tmp9, i64 0
-; IR:   %tmp11 = load volatile i32, i32 addrspace(1)* undef
-; IR:   %tmp12 = icmp slt i32 %tmp11, 9
-; IR:   %20 = xor i1 %tmp12, true
-; IR:   %21 = call i64 @llvm.amdgcn.if.break(i1 %20, i64 %phi.broken)
-; IR:   br label %Flow2
-
-; IR: bb31.loopexit:
-; IR:   br label %Flow1
+; IR: %tmp12 = icmp slt i32 %tmp11, 9
+; IR-NEXT: %27 = xor i1 %tmp12, true
+; IR-NEXT: %28 = call i64 @llvm.amdgcn.if.break(i1 %27, i64 %phi.broken)
+; IR-NEXT: br label %Flow3
+
+; IR: Flow3:
+; IR-NEXT: %loop.phi11 = phi i64 [ %phi.broken, %bb21 ], [ %phi.broken, %Flow2 ]
+; IR-NEXT: %loop.phi9 = phi i64 [ %28, %bb21 ], [ %loop.phi10, %Flow2 ]
+; IR-NEXT: %29 = phi <4 x i32> [ %tmp9, %bb21 ], [ %19, %Flow2 ]
+; IR-NEXT: %30 = phi i32 [ %tmp10, %bb21 ], [ %20, %Flow2 ]
+; IR-NEXT: %31 = phi i1 [ %27, %bb21 ], [ %21, %Flow2 ]
+; IR-NEXT: call void @llvm.amdgcn.end.cf(i64 %26)
+; IR-NEXT: br i1 %22, label %bb31.loopexit, label %Flow4
 
 ; IR: bb31:
-; IR:   call void @llvm.amdgcn.end.cf(i64 %7)
-; IR:   store volatile i32 0, i32 addrspace(1)* undef
-; IR:   ret void
+; IR-NEXT: call void @llvm.amdgcn.end.cf(i64 %7)
+; IR-NEXT: store volatile i32 0, i32 addrspace(1)* undef
+; IR-NEXT: ret void
 
 
 ; GCN-LABEL: {{^}}nested_loop_conditions:
diff --git a/test/Transforms/StructurizeCFG/AMDGPU/backedge-id-bug.ll b/test/Transforms/StructurizeCFG/AMDGPU/backedge-id-bug.ll
index 9cddffdd179..5b5ea676bae 100644
--- a/test/Transforms/StructurizeCFG/AMDGPU/backedge-id-bug.ll
+++ b/test/Transforms/StructurizeCFG/AMDGPU/backedge-id-bug.ll
@@ -1,3 +1,4 @@
+; XFAIL: *
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -structurizecfg %s | FileCheck %s
 
diff --git a/test/Transforms/StructurizeCFG/bug36015.ll b/test/Transforms/StructurizeCFG/bug36015.ll
new file mode 100644
index 00000000000..24b9c9cdde2
--- /dev/null
+++ b/test/Transforms/StructurizeCFG/bug36015.ll
@@ -0,0 +1,53 @@
+; RUN: opt -S -structurizecfg %s | FileCheck %s
+
+; r321751 introduced a bug where control flow branching from if to exit was
+; not handled properly and instead ended up in an infinite loop.
+define void @bug36015(i32 %cmp0, i32 %count) {
+entry:
+  br label %loop.outer
+
+loop.outer:
+  %ctr.loop.outer = phi i32 [ 0, %entry ], [ %ctr.else, %else ]
+  call void @foo(i32 0)
+  br label %loop.inner
+
+loop.inner:
+  %ctr.loop.inner = phi i32 [ %ctr.loop.outer, %loop.outer ], [ %ctr.if, %if ]
+  call void @foo(i32 1)
+  %cond.inner = icmp eq i32 %cmp0, %ctr.loop.inner
+  br i1 %cond.inner, label %if, label %else
+
+; CHECK: if:
+; CHECK:   %0 = xor i1 %cond.if, true
+; CHECK:   br label %Flow
+if:
+  %ctr.if = add i32 %ctr.loop.inner, 1
+  call void @foo(i32 2)
+  %cond.if = icmp slt i32 %ctr.if, %count
+  br i1 %cond.if, label %loop.inner, label %exit
+
+; CHECK: Flow:
+; CHECK:   %2 = phi i1 [ %0, %if ], [ true, %loop.inner ]
+; CHECK:   %3 = phi i1 [ false, %if ], [ true, %loop.inner ]
+; CHECK:   br i1 %2, label %Flow1, label %loop.inner
+
+; CHECK: Flow1:
+; CHECK:   br i1 %3, label %else, label %Flow2
+
+; CHECK: else:
+; CHECK:   br label %Flow2
+else:
+  %ctr.else = add i32 %ctr.loop.inner, 1
+  call void @foo(i32 3)
+  %cond.else = icmp slt i32 %ctr.else, %count
+  br i1 %cond.else, label %loop.outer, label %exit
+
+; CHECK: Flow2:
+; CHECK:   %6 = phi i1 [ %4, %else ], [ true, %Flow1 ]
+; CHECK:   br i1 %6, label %exit, label %loop.outer
+
+exit:
+  ret void
+}
+
+declare void @foo(i32)
diff --git a/test/Transforms/StructurizeCFG/nested-loop-order.ll b/test/Transforms/StructurizeCFG/nested-loop-order.ll
index 7b5bd5acb62..58634d0d37d 100644
--- a/test/Transforms/StructurizeCFG/nested-loop-order.ll
+++ b/test/Transforms/StructurizeCFG/nested-loop-order.ll
@@ -1,76 +1,32 @@
-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S -structurizecfg %s -o - | FileCheck %s
 
 define void @main(float addrspace(1)* %out) {
-; CHECK-LABEL: @main(
-; CHECK-NEXT:  main_body:
-; CHECK-NEXT:    br label [[LOOP_OUTER:%.*]]
-; CHECK:       LOOP.outer:
-; CHECK-NEXT:    [[TEMP8_0_PH:%.*]] = phi float [ 0.000000e+00, [[MAIN_BODY:%.*]] ], [ [[TMP13:%.*]], [[FLOW3:%.*]] ]
-; CHECK-NEXT:    [[TEMP4_0_PH:%.*]] = phi i32 [ 0, [[MAIN_BODY]] ], [ [[TMP12:%.*]], [[FLOW3]] ]
-; CHECK-NEXT:    br label [[LOOP:%.*]]
-; CHECK:       LOOP:
-; CHECK-NEXT:    [[TMP0:%.*]] = phi i32 [ undef, [[LOOP_OUTER]] ], [ [[TMP12]], [[FLOW:%.*]] ]
-; CHECK-NEXT:    [[TMP1:%.*]] = phi float [ undef, [[LOOP_OUTER]] ], [ [[TMP13]], [[FLOW]] ]
-; CHECK-NEXT:    [[TEMP4_0:%.*]] = phi i32 [ [[TEMP4_0_PH]], [[LOOP_OUTER]] ], [ [[TMP15:%.*]], [[FLOW]] ]
-; CHECK-NEXT:    [[TMP20:%.*]] = add i32 [[TEMP4_0]], 1
-; CHECK-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP20]], 3
-; CHECK-NEXT:    [[TMP2:%.*]] = xor i1 [[TMP22]], true
-; CHECK-NEXT:    br i1 [[TMP2]], label [[ENDIF:%.*]], label [[FLOW]]
-; CHECK:       Flow2:
-; CHECK-NEXT:    [[TMP3:%.*]] = phi float [ [[TEMP8_0_PH]], [[IF29:%.*]] ], [ [[TMP9:%.*]], [[FLOW1:%.*]] ]
-; CHECK-NEXT:    [[TMP4:%.*]] = phi i32 [ [[TMP20]], [[IF29]] ], [ undef, [[FLOW1]] ]
-; CHECK-NEXT:    [[TMP5:%.*]] = phi i1 [ [[TMP32:%.*]], [[IF29]] ], [ true, [[FLOW1]] ]
-; CHECK-NEXT:    br label [[FLOW]]
-; CHECK:       Flow3:
-; CHECK-NEXT:    br i1 [[TMP16:%.*]], label [[ENDLOOP:%.*]], label [[LOOP_OUTER]]
-; CHECK:       ENDLOOP:
-; CHECK-NEXT:    [[TEMP8_1:%.*]] = phi float [ [[TMP14:%.*]], [[FLOW3]] ]
-; CHECK-NEXT:    [[TMP23:%.*]] = icmp eq i32 [[TMP20]], 3
-; CHECK-NEXT:    [[DOT45:%.*]] = select i1 [[TMP23]], float 0.000000e+00, float 1.000000e+00
-; CHECK-NEXT:    store float [[DOT45]], float addrspace(1)* [[OUT:%.*]]
-; CHECK-NEXT:    ret void
-; CHECK:       ENDIF:
-; CHECK-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP20]], 1
-; CHECK-NEXT:    [[TMP6:%.*]] = xor i1 [[TMP31]], true
-; CHECK-NEXT:    br i1 [[TMP6]], label [[ENDIF28:%.*]], label [[FLOW1]]
-; CHECK:       Flow1:
-; CHECK-NEXT:    [[TMP7:%.*]] = phi i32 [ [[TMP20]], [[ENDIF28]] ], [ [[TMP0]], [[ENDIF]] ]
-; CHECK-NEXT:    [[TMP8:%.*]] = phi float [ [[TMP35:%.*]], [[ENDIF28]] ], [ [[TMP1]], [[ENDIF]] ]
-; CHECK-NEXT:    [[TMP9]] = phi float [ [[TMP35]], [[ENDIF28]] ], [ [[TEMP8_0_PH]], [[ENDIF]] ]
-; CHECK-NEXT:    [[TMP10:%.*]] = phi i1 [ [[TMP36:%.*]], [[ENDIF28]] ], [ true, [[ENDIF]] ]
-; CHECK-NEXT:    [[TMP11:%.*]] = phi i1 [ false, [[ENDIF28]] ], [ true, [[ENDIF]] ]
-; CHECK-NEXT:    br i1 [[TMP11]], label [[IF29]], label [[FLOW2:%.*]]
-; CHECK:       IF29:
-; CHECK-NEXT:    [[TMP32]] = icmp sgt i32 [[TMP20]], 2
-; CHECK-NEXT:    br label [[FLOW2]]
-; CHECK:       Flow:
-; CHECK-NEXT:    [[TMP12]] = phi i32 [ [[TMP7]], [[FLOW2]] ], [ [[TMP0]], [[LOOP]] ]
-; CHECK-NEXT:    [[TMP13]] = phi float [ [[TMP8]], [[FLOW2]] ], [ [[TMP1]], [[LOOP]] ]
-; CHECK-NEXT:    [[TMP14]] = phi float [ [[TMP3]], [[FLOW2]] ], [ [[TEMP8_0_PH]], [[LOOP]] ]
-; CHECK-NEXT:    [[TMP15]] = phi i32 [ [[TMP4]], [[FLOW2]] ], [ undef, [[LOOP]] ]
-; CHECK-NEXT:    [[TMP16]] = phi i1 [ [[TMP10]], [[FLOW2]] ], [ true, [[LOOP]] ]
-; CHECK-NEXT:    [[TMP17:%.*]] = phi i1 [ [[TMP5]], [[FLOW2]] ], [ true, [[LOOP]] ]
-; CHECK-NEXT:    br i1 [[TMP17]], label [[FLOW3]], label [[LOOP]]
-; CHECK:       ENDIF28:
-; CHECK-NEXT:    [[TMP35]] = fadd float [[TEMP8_0_PH]], 1.000000e+00
-; CHECK-NEXT:    [[TMP36]] = icmp sgt i32 [[TMP20]], 2
-; CHECK-NEXT:    br label [[FLOW1]]
-;
+
+; CHECK: main_body:
+; CHECK: br label %LOOP.outer
 main_body:
   br label %LOOP.outer
 
+; CHECK: LOOP.outer:
+; CHECK: br label %LOOP
 LOOP.outer:                                       ; preds = %ENDIF28, %main_body
   %temp8.0.ph = phi float [ 0.000000e+00, %main_body ], [ %tmp35, %ENDIF28 ]
   %temp4.0.ph = phi i32 [ 0, %main_body ], [ %tmp20, %ENDIF28 ]
   br label %LOOP
 
+; CHECK: LOOP:
+; br i1 %{{[0-9]+}}, label %ENDIF, label %Flow
 LOOP:                                             ; preds = %IF29, %LOOP.outer
   %temp4.0 = phi i32 [ %temp4.0.ph, %LOOP.outer ], [ %tmp20, %IF29 ]
   %tmp20 = add i32 %temp4.0, 1
   %tmp22 = icmp sgt i32 %tmp20, 3
   br i1 %tmp22, label %ENDLOOP, label %ENDIF
 
+; CHECK: Flow3
+; CHECK: br i1 %{{[0-9]+}}, label %ENDLOOP, label %LOOP.outer
+
+; CHECK: ENDLOOP:
+; CHECK: ret void
 ENDLOOP:                                          ; preds = %ENDIF28, %IF29, %LOOP
   %temp8.1 = phi float [ %temp8.0.ph, %LOOP ], [ %temp8.0.ph, %IF29 ], [ %tmp35, %ENDIF28 ]
   %tmp23 = icmp eq i32 %tmp20, 3
@@ -78,14 +34,29 @@ ENDLOOP:                                          ; preds = %ENDIF28, %IF29, %LO
   store float %.45, float addrspace(1)* %out
   ret void
 
+; CHECK: ENDIF:
+; CHECK: br i1 %tmp31, label %IF29, label %Flow1
 ENDIF:                                            ; preds = %LOOP
   %tmp31 = icmp sgt i32 %tmp20, 1
   br i1 %tmp31, label %IF29, label %ENDIF28
 
+; CHECK: Flow:
+; CHECK: br i1 %{{[0-9]+}}, label %Flow2, label %LOOP
+
+; CHECK: IF29:
+; CHECK: br label %Flow1
 IF29:                                             ; preds = %ENDIF
   %tmp32 = icmp sgt i32 %tmp20, 2
   br i1 %tmp32, label %ENDLOOP, label %LOOP
 
+; CHECK: Flow1:
+; CHECK: br label %Flow
+
+; CHECK: Flow2:
+; CHECK: br i1 %{{[0-9]+}}, label %ENDIF28, label %Flow3
+
+; CHECK: ENDIF28:
+; CHECK: br label %Flow3
 ENDIF28:                                          ; preds = %ENDIF
   %tmp35 = fadd float %temp8.0.ph, 1.0
   %tmp36 = icmp sgt i32 %tmp20, 2