StructurizeCFG: Directly invert cmp instructions
The most common case for a branch condition is
a single use compare. Directly invert the branch
predicate rather than adding a lot of xor i1 true
which the DAG will have to fold later.
This produces nicer to read structurizer output.
This produces some random changes in codegen
due to the DAG swapping branch conditions itself,
and then does a poor job of dealing with those
inverts.
llvm-svn: 300732
diff --git a/llvm/test/Transforms/StructurizeCFG/invert-compare.ll b/llvm/test/Transforms/StructurizeCFG/invert-compare.ll
new file mode 100644
index 0000000..87d9c6d
--- /dev/null
+++ b/llvm/test/Transforms/StructurizeCFG/invert-compare.ll
@@ -0,0 +1,60 @@
+; RUN: opt -S -structurizecfg %s | FileCheck %s
+
+; CHECK-LABEL: @directly_invert_compare_condition_jump_into_loop(
+; CHECK: %cmp0 = fcmp uge float %arg0, %arg1
+; CHECK-NEXT: br i1 %cmp0, label %end.loop, label %Flow
+define void @directly_invert_compare_condition_jump_into_loop(i32 addrspace(1)* %out, i32 %n, float %arg0, float %arg1) #0 {
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i32 [0, %entry], [%i.inc, %end.loop]
+ %ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %i
+ store i32 %i, i32 addrspace(1)* %ptr, align 4
+ %cmp0 = fcmp olt float %arg0, %arg1
+ br i1 %cmp0, label %mid.loop, label %end.loop
+
+mid.loop:
+ store i32 333, i32 addrspace(1)* %out, align 4
+ br label %for.end
+
+end.loop:
+ %i.inc = add i32 %i, 1
+ %cmp = icmp ne i32 %i.inc, %n
+ br i1 %cmp, label %for.body, label %for.end
+
+for.end:
+ ret void
+}
+
+; CHECK-LABEL: @invert_multi_use_compare_condition_jump_into_loop(
+; CHECK: %cmp0 = fcmp olt float %arg0, %arg1
+; CHECK: store volatile i1 %cmp0, i1 addrspace(1)* undef
+; CHECK: %0 = xor i1 %cmp0, true
+; CHECK-NEXT: br i1 %0, label %end.loop, label %Flow
+define void @invert_multi_use_compare_condition_jump_into_loop(i32 addrspace(1)* %out, i32 %n, float %arg0, float %arg1) #0 {
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i32 [0, %entry], [%i.inc, %end.loop]
+ %ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %i
+ store i32 %i, i32 addrspace(1)* %ptr, align 4
+ %cmp0 = fcmp olt float %arg0, %arg1
+ store volatile i1 %cmp0, i1 addrspace(1)* undef
+ br i1 %cmp0, label %mid.loop, label %end.loop
+
+mid.loop:
+ store i32 333, i32 addrspace(1)* %out, align 4
+ br label %for.end
+
+end.loop:
+ %i.inc = add i32 %i, 1
+ %cmp = icmp ne i32 %i.inc, %n
+ br i1 %cmp, label %for.body, label %for.end
+
+for.end:
+ ret void
+}
+
+attributes #0 = { nounwind }
\ No newline at end of file
diff --git a/llvm/test/Transforms/StructurizeCFG/one-loop-multiple-backedges.ll b/llvm/test/Transforms/StructurizeCFG/one-loop-multiple-backedges.ll
index 668a1e9..aff5964 100644
--- a/llvm/test/Transforms/StructurizeCFG/one-loop-multiple-backedges.ll
+++ b/llvm/test/Transforms/StructurizeCFG/one-loop-multiple-backedges.ll
@@ -11,8 +11,8 @@
bb3: ; preds = %bb7, %bb
%tmp = phi i64 [ 0, %bb ], [ %tmp8, %bb7 ]
%tmp4 = fcmp ult float %arg1, 3.500000e+00
-; CHECK: %0 = xor i1 %tmp4, true
-; CHECK: br i1 %0, label %bb5, label %Flow
+; CHECK: %tmp4 = fcmp oge float %arg1, 3.500000e+00
+; CHECK: br i1 %tmp4, label %bb5, label %Flow
br i1 %tmp4, label %bb7, label %bb5
; CHECK: bb5:
@@ -22,7 +22,8 @@
br i1 %tmp6, label %bb10, label %bb7
; CHECK: Flow:
-; CHECK: br i1 %3, label %bb7, label %Flow1
+; CHECK: %1 = phi i1 [ %tmp6, %bb5 ], [ %tmp4, %bb3 ]
+; CHECK-NEXT: br i1 %1, label %bb7, label %Flow1
; CHECK: bb7
bb7: ; preds = %bb5, %bb3
@@ -32,9 +33,10 @@
br i1 %tmp9, label %bb3, label %bb10
; CHECK: Flow1:
-; CHECK: br i1 %7, label %bb10, label %bb3
+; CHECK: %4 = phi i1 [ %tmp9, %bb7 ], [ true, %Flow ]
+; CHECK-NEXT: br i1 %4, label %bb10, label %bb3
-; CHECK: bb10
+; CHECK: bb10:
bb10: ; preds = %bb7, %bb5
%tmp11 = phi i32 [ 15, %bb5 ], [ 255, %bb7 ]
store i32 %tmp11, i32 addrspace(1)* %arg, align 4
diff --git a/llvm/test/Transforms/StructurizeCFG/post-order-traversal-bug.ll b/llvm/test/Transforms/StructurizeCFG/post-order-traversal-bug.ll
index ba9aa29..a8835f1 100644
--- a/llvm/test/Transforms/StructurizeCFG/post-order-traversal-bug.ll
+++ b/llvm/test/Transforms/StructurizeCFG/post-order-traversal-bug.ll
@@ -59,7 +59,8 @@
; CHECK: br i1 %{{[0-9]}}, label %for.body.1, label %Flow2
; CHECK: for.body.1:
-; CHECK: br i1 %{{[0-9]+}}, label %for.body.6, label %Flow3
+; CHECK: %cmp1.5 = icmp ne i32 %tmp22, %K1
+; CHECK-NEXT: br i1 %cmp1.5, label %for.body.6, label %Flow3
for.body.1: ; preds = %if.then, %lor.lhs.false
%best_val.233 = phi float [ %tmp5, %if.then ], [ %best_val.027, %lor.lhs.false ]
%best_count.231 = phi i32 [ %sub4, %if.then ], [ %best_count.025, %lor.lhs.false ]