summaryrefslogtreecommitdiffstats
path: root/llvm/test/Transforms/CodeGenPrepare/X86/select.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/Transforms/CodeGenPrepare/X86/select.ll')
-rw-r--r--llvm/test/Transforms/CodeGenPrepare/X86/select.ll205
1 files changed, 205 insertions, 0 deletions
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/select.ll b/llvm/test/Transforms/CodeGenPrepare/X86/select.ll
new file mode 100644
index 00000000000..7829376e9db
--- /dev/null
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/select.ll
@@ -0,0 +1,205 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -codegenprepare -S < %s | FileCheck %s
+; RUN: opt -debugify -codegenprepare -S < %s | FileCheck %s -check-prefix=DEBUG
+
+target triple = "x86_64-unknown-unknown"
+
+; Nothing to sink and convert here.
+
+define i32 @no_sink(double %a, double* %b, i32 %x, i32 %y) {
+; CHECK-LABEL: @no_sink(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[LOAD:%.*]] = load double, double* [[B:%.*]], align 8
+; CHECK-NEXT: [[CMP:%.*]] = fcmp olt double [[LOAD]], [[A:%.*]]
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], i32 [[X:%.*]], i32 [[Y:%.*]]
+; CHECK-NEXT: ret i32 [[SEL]]
+;
+entry:
+ %load = load double, double* %b, align 8
+ %cmp = fcmp olt double %load, %a
+ %sel = select i1 %cmp, i32 %x, i32 %y
+ ret i32 %sel
+}
+
+
+; An 'fdiv' is expensive, so sink it rather than speculatively execute it.
+
+define float @fdiv_true_sink(float %a, float %b) {
+; CHECK-LABEL: @fdiv_true_sink(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt float [[A:%.*]], 1.000000e+00
+; CHECK-NEXT: br i1 [[CMP]], label [[SELECT_TRUE_SINK:%.*]], label [[SELECT_END:%.*]]
+; CHECK: select.true.sink:
+; CHECK-NEXT: [[DIV:%.*]] = fdiv float [[A]], [[B:%.*]]
+; CHECK-NEXT: br label [[SELECT_END]]
+; CHECK: select.end:
+; CHECK-NEXT: [[SEL:%.*]] = phi float [ [[DIV]], [[SELECT_TRUE_SINK]] ], [ 2.000000e+00, [[ENTRY:%.*]] ]
+; CHECK-NEXT: ret float [[SEL]]
+;
+; DEBUG-LABEL: @fdiv_true_sink(
+; DEBUG-NEXT: entry:
+; DEBUG-NEXT: [[CMP:%.*]] = fcmp ogt float [[A:%.*]], 1.000000e+00
+; DEBUG-NEXT: call void @llvm.dbg.value(metadata i1 [[CMP]]
+; DEBUG-NEXT: br i1 [[CMP]], label [[SELECT_TRUE_SINK:%.*]], label [[SELECT_END:%.*]], !dbg
+; DEBUG: select.true.sink:
+; DEBUG-NEXT: [[DIV:%.*]] = fdiv float [[A]], [[B:%.*]]
+; DEBUG-NEXT: call void @llvm.dbg.value(metadata float [[DIV]]
+; DEBUG-NEXT: br label [[SELECT_END]], !dbg
+; DEBUG: select.end:
+; DEBUG-NEXT: [[SEL:%.*]] = phi float [ [[DIV]], [[SELECT_TRUE_SINK]] ], [ 2.000000e+00, [[ENTRY:%.*]] ], !dbg
+; DEBUG-NEXT: call void @llvm.dbg.value(metadata float [[SEL]]
+; DEBUG-NEXT: ret float [[SEL]]
+;
+entry:
+ %div = fdiv float %a, %b
+ %cmp = fcmp ogt float %a, 1.0
+ %sel = select i1 %cmp, float %div, float 2.0
+ ret float %sel
+}
+
+define float @fdiv_false_sink(float %a, float %b) {
+; CHECK-LABEL: @fdiv_false_sink(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt float [[A:%.*]], 3.000000e+00
+; CHECK-NEXT: br i1 [[CMP]], label [[SELECT_END:%.*]], label [[SELECT_FALSE_SINK:%.*]]
+; CHECK: select.false.sink:
+; CHECK-NEXT: [[DIV:%.*]] = fdiv float [[A]], [[B:%.*]]
+; CHECK-NEXT: br label [[SELECT_END]]
+; CHECK: select.end:
+; CHECK-NEXT: [[SEL:%.*]] = phi float [ 4.000000e+00, [[ENTRY:%.*]] ], [ [[DIV]], [[SELECT_FALSE_SINK]] ]
+; CHECK-NEXT: ret float [[SEL]]
+;
+; DEBUG-LABEL: @fdiv_false_sink(
+; DEBUG-NEXT: entry:
+; DEBUG-NEXT: [[CMP:%.*]] = fcmp ogt float [[A:%.*]], 3.000000e+00
+; DEBUG-NEXT: call void @llvm.dbg.value(metadata i1 [[CMP]]
+; DEBUG-NEXT: br i1 [[CMP]], label [[SELECT_END:%.*]], label [[SELECT_FALSE_SINK:%.*]], !dbg
+; DEBUG: select.false.sink:
+; DEBUG-NEXT: [[DIV:%.*]] = fdiv float [[A]], [[B:%.*]]
+; DEBUG-NEXT: call void @llvm.dbg.value(metadata float [[DIV]]
+; DEBUG-NEXT: br label [[SELECT_END]], !dbg
+; DEBUG: select.end:
+; DEBUG-NEXT: [[SEL:%.*]] = phi float [ 4.000000e+00, [[ENTRY:%.*]] ], [ [[DIV]], [[SELECT_FALSE_SINK]] ], !dbg
+; DEBUG-NEXT: call void @llvm.dbg.value(metadata float [[SEL]]
+; DEBUG-NEXT: ret float [[SEL]], !dbg
+;
+entry:
+ %div = fdiv float %a, %b
+ %cmp = fcmp ogt float %a, 3.0
+ %sel = select i1 %cmp, float 4.0, float %div
+ ret float %sel
+}
+
+define float @fdiv_both_sink(float %a, float %b) {
+; CHECK-LABEL: @fdiv_both_sink(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt float [[A:%.*]], 5.000000e+00
+; CHECK-NEXT: br i1 [[CMP]], label [[SELECT_TRUE_SINK:%.*]], label [[SELECT_FALSE_SINK:%.*]]
+; CHECK: select.true.sink:
+; CHECK-NEXT: [[DIV1:%.*]] = fdiv float [[A]], [[B:%.*]]
+; CHECK-NEXT: br label [[SELECT_END:%.*]]
+; CHECK: select.false.sink:
+; CHECK-NEXT: [[DIV2:%.*]] = fdiv float [[B]], [[A]]
+; CHECK-NEXT: br label [[SELECT_END]]
+; CHECK: select.end:
+; CHECK-NEXT: [[SEL:%.*]] = phi float [ [[DIV1]], [[SELECT_TRUE_SINK]] ], [ [[DIV2]], [[SELECT_FALSE_SINK]] ]
+; CHECK-NEXT: ret float [[SEL]]
+;
+entry:
+ %div1 = fdiv float %a, %b
+ %div2 = fdiv float %b, %a
+ %cmp = fcmp ogt float %a, 5.0
+ %sel = select i1 %cmp, float %div1, float %div2
+ ret float %sel
+}
+
+; But if the select is marked unpredictable, then don't turn it into a branch.
+
+define float @unpredictable_select(float %a, float %b) {
+; CHECK-LABEL: @unpredictable_select(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[DIV:%.*]] = fdiv float [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt float [[A]], 1.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], float [[DIV]], float 2.000000e+00, !unpredictable !0
+; CHECK-NEXT: ret float [[SEL]]
+;
+entry:
+ %div = fdiv float %a, %b
+ %cmp = fcmp ogt float %a, 1.0
+ %sel = select i1 %cmp, float %div, float 2.0, !unpredictable !0
+ ret float %sel
+}
+
+!0 = !{}
+
+; An 'fadd' is not too expensive, so it's ok to speculate.
+
+define float @fadd_no_sink(float %a, float %b) {
+; CHECK-LABEL: @fadd_no_sink(
+; CHECK-NEXT: [[ADD:%.*]] = fadd float [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt float 6.000000e+00, [[A]]
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], float [[ADD]], float 7.000000e+00
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %add = fadd float %a, %b
+ %cmp = fcmp ogt float 6.0, %a
+ %sel = select i1 %cmp, float %add, float 7.0
+ ret float %sel
+}
+
+; Possible enhancement: sinkability is only calculated with the direct
+; operand of the select, so we don't try to sink this. The fdiv cost is not
+; taken into account.
+
+define float @fdiv_no_sink(float %a, float %b) {
+; CHECK-LABEL: @fdiv_no_sink(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[DIV:%.*]] = fdiv float [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[ADD:%.*]] = fadd float [[DIV]], [[B]]
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt float [[A]], 1.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], float [[ADD]], float 8.000000e+00
+; CHECK-NEXT: ret float [[SEL]]
+;
+entry:
+ %div = fdiv float %a, %b
+ %add = fadd float %div, %b
+ %cmp = fcmp ogt float %a, 1.0
+ %sel = select i1 %cmp, float %add, float 8.0
+ ret float %sel
+}
+
+; Do not transform the CFG if the select operands may have side effects.
+
+declare i64* @bar(i32, i32, i32)
+declare i64* @baz(i32, i32, i32)
+
+define i64* @calls_no_sink(i32 %in) {
+; CHECK-LABEL: @calls_no_sink(
+; CHECK-NEXT: [[CALL1:%.*]] = call i64* @bar(i32 1, i32 2, i32 3)
+; CHECK-NEXT: [[CALL2:%.*]] = call i64* @baz(i32 1, i32 2, i32 3)
+; CHECK-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[IN:%.*]], 0
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[TOBOOL]], i64* [[CALL1]], i64* [[CALL2]]
+; CHECK-NEXT: ret i64* [[SEL]]
+;
+ %call1 = call i64* @bar(i32 1, i32 2, i32 3)
+ %call2 = call i64* @baz(i32 1, i32 2, i32 3)
+ %tobool = icmp ne i32 %in, 0
+ %sel = select i1 %tobool, i64* %call1, i64* %call2
+ ret i64* %sel
+}
+
+define i32 @sdiv_no_sink(i32 %a, i32 %b) {
+; CHECK-LABEL: @sdiv_no_sink(
+; CHECK-NEXT: [[DIV1:%.*]] = sdiv i32 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[DIV2:%.*]] = sdiv i32 [[B]], [[A]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[A]], 5
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], i32 [[DIV1]], i32 [[DIV2]]
+; CHECK-NEXT: ret i32 [[SEL]]
+;
+ %div1 = sdiv i32 %a, %b
+ %div2 = sdiv i32 %b, %a
+ %cmp = icmp sgt i32 %a, 5
+ %sel = select i1 %cmp, i32 %div1, i32 %div2
+ ret i32 %sel
+}
+
OpenPOWER on IntegriCloud