Don't use a potentially expensive shift if all we want is one set bit.

[oota-llvm.git] / test / CodeGen / SystemZ / int-sub-03.ll

diff --git a/test/CodeGen/SystemZ/int-sub-03.ll b/test/CodeGen/SystemZ/int-sub-03.ll
index 73571b3591f5971a5b3a833319f7c7151b777e99..49040177fde7d0490b79ad05b804a469a32f3b12 100644 (file)
--- a/test/CodeGen/SystemZ/int-sub-03.ll
+++ b/test/CodeGen/SystemZ/int-sub-03.ll
@@ -2,6 +2,8 @@
 ;
 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
 
+declare i64 @foo()
+
 ; Check SLGFR.
 define i64 @f1(i64 %a, i32 %b) {
 ; CHECK: f1:
@@ -100,3 +102,79 @@ define i64 @f8(i64 %a, i64 %src, i64 %index) {
   %sub = sub i64 %a, %bext
   ret i64 %sub
 }
+
+; Check that subtractions of spilled values can use SLGF rather than SLGFR.
+define i64 @f9(i32 *%ptr0) {
+; CHECK: f9:
+; CHECK: brasl %r14, foo@PLT
+; CHECK: slgf %r2, 16{{[04]}}(%r15)
+; CHECK: br %r14
+  %ptr1 = getelementptr i32 *%ptr0, i64 2
+  %ptr2 = getelementptr i32 *%ptr0, i64 4
+  %ptr3 = getelementptr i32 *%ptr0, i64 6
+  %ptr4 = getelementptr i32 *%ptr0, i64 8
+  %ptr5 = getelementptr i32 *%ptr0, i64 10
+  %ptr6 = getelementptr i32 *%ptr0, i64 12
+  %ptr7 = getelementptr i32 *%ptr0, i64 14
+  %ptr8 = getelementptr i32 *%ptr0, i64 16
+  %ptr9 = getelementptr i32 *%ptr0, i64 18
+
+  %val0 = load i32 *%ptr0
+  %val1 = load i32 *%ptr1
+  %val2 = load i32 *%ptr2
+  %val3 = load i32 *%ptr3
+  %val4 = load i32 *%ptr4
+  %val5 = load i32 *%ptr5
+  %val6 = load i32 *%ptr6
+  %val7 = load i32 *%ptr7
+  %val8 = load i32 *%ptr8
+  %val9 = load i32 *%ptr9
+
+  %frob0 = add i32 %val0, 100
+  %frob1 = add i32 %val1, 100
+  %frob2 = add i32 %val2, 100
+  %frob3 = add i32 %val3, 100
+  %frob4 = add i32 %val4, 100
+  %frob5 = add i32 %val5, 100
+  %frob6 = add i32 %val6, 100
+  %frob7 = add i32 %val7, 100
+  %frob8 = add i32 %val8, 100
+  %frob9 = add i32 %val9, 100
+
+  store i32 %frob0, i32 *%ptr0
+  store i32 %frob1, i32 *%ptr1
+  store i32 %frob2, i32 *%ptr2
+  store i32 %frob3, i32 *%ptr3
+  store i32 %frob4, i32 *%ptr4
+  store i32 %frob5, i32 *%ptr5
+  store i32 %frob6, i32 *%ptr6
+  store i32 %frob7, i32 *%ptr7
+  store i32 %frob8, i32 *%ptr8
+  store i32 %frob9, i32 *%ptr9
+
+  %ret = call i64 @foo()
+
+  %ext0 = zext i32 %frob0 to i64
+  %ext1 = zext i32 %frob1 to i64
+  %ext2 = zext i32 %frob2 to i64
+  %ext3 = zext i32 %frob3 to i64
+  %ext4 = zext i32 %frob4 to i64
+  %ext5 = zext i32 %frob5 to i64
+  %ext6 = zext i32 %frob6 to i64
+  %ext7 = zext i32 %frob7 to i64
+  %ext8 = zext i32 %frob8 to i64
+  %ext9 = zext i32 %frob9 to i64
+
+  %sub0 = sub i64 %ret, %ext0
+  %sub1 = sub i64 %sub0, %ext1
+  %sub2 = sub i64 %sub1, %ext2
+  %sub3 = sub i64 %sub2, %ext3
+  %sub4 = sub i64 %sub3, %ext4
+  %sub5 = sub i64 %sub4, %ext5
+  %sub6 = sub i64 %sub5, %ext6
+  %sub7 = sub i64 %sub6, %ext7
+  %sub8 = sub i64 %sub7, %ext8
+  %sub9 = sub i64 %sub8, %ext9
+
+  ret i64 %sub9
+}