[SystemZ] Optimize sign-extends of vector setccs
Normal (sext (setcc ...)) sequences are optimised into
(select_cc ..., -1, 0) by DAGCombiner::visitSIGN_EXTEND.
However, this is deliberately not done for vectors, and after
vector type legalization we have (sext_inreg (setcc ...)) instead.
I wondered about trying to extend DAGCombiner to handle this case too,
but it seemed to be a loss on some other targets I tried, even those for
which SETCC isn't "legal" and SELECT_CC is.
llvm-svn: 186149
diff --git a/llvm/test/CodeGen/SystemZ/branch-07.ll b/llvm/test/CodeGen/SystemZ/branch-07.ll
index 00e8b8a..d009af2 100644
--- a/llvm/test/CodeGen/SystemZ/branch-07.ll
+++ b/llvm/test/CodeGen/SystemZ/branch-07.ll
@@ -4,6 +4,7 @@
declare i64 @foo()
+; Test EQ.
define void @f1(i64 %target) {
; CHECK: f1:
; CHECK: .cfi_def_cfa_offset
@@ -18,6 +19,7 @@
ret void
}
+; Test NE.
define void @f2(i64 %target) {
; CHECK: f2:
; CHECK: .cfi_def_cfa_offset
@@ -32,6 +34,7 @@
ret void
}
+; Test SLE.
define void @f3(i64 %target) {
; CHECK: f3:
; CHECK: .cfi_def_cfa_offset
@@ -46,6 +49,7 @@
ret void
}
+; Test SLT.
define void @f4(i64 %target) {
; CHECK: f4:
; CHECK: .cfi_def_cfa_offset
@@ -60,6 +64,7 @@
ret void
}
+; Test SGT.
define void @f5(i64 %target) {
; CHECK: f5:
; CHECK: .cfi_def_cfa_offset
@@ -74,6 +79,7 @@
ret void
}
+; Test SGE.
define void @f6(i64 %target) {
; CHECK: f6:
; CHECK: .cfi_def_cfa_offset
@@ -87,3 +93,67 @@
exit:
ret void
}
+
+; Test a vector of 0/-1 results for i32 EQ.
+define i64 @f7(i64 %a, i64 %b) {
+; CHECK: f7:
+; CHECK: lhi [[REG:%r[0-5]]], -1
+; CHECK: crje {{%r[0-5]}}
+; CHECK: lhi [[REG]], 0
+; CHECK-NOT: sra
+; CHECK: br %r14
+ %avec = bitcast i64 %a to <2 x i32>
+ %bvec = bitcast i64 %b to <2 x i32>
+ %cmp = icmp eq <2 x i32> %avec, %bvec
+ %ext = sext <2 x i1> %cmp to <2 x i32>
+ %ret = bitcast <2 x i32> %ext to i64
+ ret i64 %ret
+}
+
+; Test a vector of 0/-1 results for i32 NE.
+define i64 @f8(i64 %a, i64 %b) {
+; CHECK: f8:
+; CHECK: lhi [[REG:%r[0-5]]], -1
+; CHECK: crjlh {{%r[0-5]}}
+; CHECK: lhi [[REG]], 0
+; CHECK-NOT: sra
+; CHECK: br %r14
+ %avec = bitcast i64 %a to <2 x i32>
+ %bvec = bitcast i64 %b to <2 x i32>
+ %cmp = icmp ne <2 x i32> %avec, %bvec
+ %ext = sext <2 x i1> %cmp to <2 x i32>
+ %ret = bitcast <2 x i32> %ext to i64
+ ret i64 %ret
+}
+
+; Test a vector of 0/-1 results for i64 EQ.
+define void @f9(i64 %a, i64 %b, <2 x i64> *%dest) {
+; CHECK: f9:
+; CHECK: lghi [[REG:%r[0-5]]], -1
+; CHECK: crje {{%r[0-5]}}
+; CHECK: lghi [[REG]], 0
+; CHECK-NOT: sra
+; CHECK: br %r14
+ %avec = bitcast i64 %a to <2 x i32>
+ %bvec = bitcast i64 %b to <2 x i32>
+ %cmp = icmp eq <2 x i32> %avec, %bvec
+ %ext = sext <2 x i1> %cmp to <2 x i64>
+ store <2 x i64> %ext, <2 x i64> *%dest
+ ret void
+}
+
+; Test a vector of 0/-1 results for i64 NE.
+define void @f10(i64 %a, i64 %b, <2 x i64> *%dest) {
+; CHECK: f10:
+; CHECK: lghi [[REG:%r[0-5]]], -1
+; CHECK: crjlh {{%r[0-5]}}
+; CHECK: lghi [[REG]], 0
+; CHECK-NOT: sra
+; CHECK: br %r14
+ %avec = bitcast i64 %a to <2 x i32>
+ %bvec = bitcast i64 %b to <2 x i32>
+ %cmp = icmp ne <2 x i32> %avec, %bvec
+ %ext = sext <2 x i1> %cmp to <2 x i64>
+ store <2 x i64> %ext, <2 x i64> *%dest
+ ret void
+}