summaryrefslogtreecommitdiffstats
path: root/test/CodeGen/ARM/vsra.ll
diff options
context:
space:
mode:
Diffstat (limited to 'test/CodeGen/ARM/vsra.ll')
-rw-r--r--test/CodeGen/ARM/vsra.ll82
1 files changed, 65 insertions, 17 deletions
diff --git a/test/CodeGen/ARM/vsra.ll b/test/CodeGen/ARM/vsra.ll
index e2829dc..acb672d 100644
--- a/test/CodeGen/ARM/vsra.ll
+++ b/test/CodeGen/ARM/vsra.ll
@@ -1,22 +1,8 @@
-; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
-; RUN: grep {vsra\\.s8} %t | count 2
-; RUN: grep {vsra\\.s16} %t | count 2
-; RUN: grep {vsra\\.s32} %t | count 2
-; RUN: grep {vsra\\.s64} %t | count 2
-; RUN: grep {vsra\\.u8} %t | count 2
-; RUN: grep {vsra\\.u16} %t | count 2
-; RUN: grep {vsra\\.u32} %t | count 2
-; RUN: grep {vsra\\.u64} %t | count 2
-; RUN: grep {vrsra\\.s8} %t | count 2
-; RUN: grep {vrsra\\.s16} %t | count 2
-; RUN: grep {vrsra\\.s32} %t | count 2
-; RUN: grep {vrsra\\.s64} %t | count 2
-; RUN: grep {vrsra\\.u8} %t | count 2
-; RUN: grep {vrsra\\.u16} %t | count 2
-; RUN: grep {vrsra\\.u32} %t | count 2
-; RUN: grep {vrsra\\.u64} %t | count 2
+; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
define <8 x i8> @vsras8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+;CHECK: vsras8:
+;CHECK: vsra.s8
%tmp1 = load <8 x i8>* %A
%tmp2 = load <8 x i8>* %B
%tmp3 = ashr <8 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
@@ -25,6 +11,8 @@ define <8 x i8> @vsras8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
}
define <4 x i16> @vsras16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+;CHECK: vsras16:
+;CHECK: vsra.s16
%tmp1 = load <4 x i16>* %A
%tmp2 = load <4 x i16>* %B
%tmp3 = ashr <4 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16 >
@@ -33,6 +21,8 @@ define <4 x i16> @vsras16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
}
define <2 x i32> @vsras32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+;CHECK: vsras32:
+;CHECK: vsra.s32
%tmp1 = load <2 x i32>* %A
%tmp2 = load <2 x i32>* %B
%tmp3 = ashr <2 x i32> %tmp2, < i32 32, i32 32 >
@@ -41,6 +31,8 @@ define <2 x i32> @vsras32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
}
define <1 x i64> @vsras64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+;CHECK: vsras64:
+;CHECK: vsra.s64
%tmp1 = load <1 x i64>* %A
%tmp2 = load <1 x i64>* %B
%tmp3 = ashr <1 x i64> %tmp2, < i64 64 >
@@ -49,6 +41,8 @@ define <1 x i64> @vsras64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
}
define <16 x i8> @vsraQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+;CHECK: vsraQs8:
+;CHECK: vsra.s8
%tmp1 = load <16 x i8>* %A
%tmp2 = load <16 x i8>* %B
%tmp3 = ashr <16 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
@@ -57,6 +51,8 @@ define <16 x i8> @vsraQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
}
define <8 x i16> @vsraQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+;CHECK: vsraQs16:
+;CHECK: vsra.s16
%tmp1 = load <8 x i16>* %A
%tmp2 = load <8 x i16>* %B
%tmp3 = ashr <8 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16 >
@@ -65,6 +61,8 @@ define <8 x i16> @vsraQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
}
define <4 x i32> @vsraQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+;CHECK: vsraQs32:
+;CHECK: vsra.s32
%tmp1 = load <4 x i32>* %A
%tmp2 = load <4 x i32>* %B
%tmp3 = ashr <4 x i32> %tmp2, < i32 32, i32 32, i32 32, i32 32 >
@@ -73,6 +71,8 @@ define <4 x i32> @vsraQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
}
define <2 x i64> @vsraQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+;CHECK: vsraQs64:
+;CHECK: vsra.s64
%tmp1 = load <2 x i64>* %A
%tmp2 = load <2 x i64>* %B
%tmp3 = ashr <2 x i64> %tmp2, < i64 64, i64 64 >
@@ -81,6 +81,8 @@ define <2 x i64> @vsraQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
}
define <8 x i8> @vsrau8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+;CHECK: vsrau8:
+;CHECK: vsra.u8
%tmp1 = load <8 x i8>* %A
%tmp2 = load <8 x i8>* %B
%tmp3 = lshr <8 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
@@ -89,6 +91,8 @@ define <8 x i8> @vsrau8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
}
define <4 x i16> @vsrau16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+;CHECK: vsrau16:
+;CHECK: vsra.u16
%tmp1 = load <4 x i16>* %A
%tmp2 = load <4 x i16>* %B
%tmp3 = lshr <4 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16 >
@@ -97,6 +101,8 @@ define <4 x i16> @vsrau16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
}
define <2 x i32> @vsrau32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+;CHECK: vsrau32:
+;CHECK: vsra.u32
%tmp1 = load <2 x i32>* %A
%tmp2 = load <2 x i32>* %B
%tmp3 = lshr <2 x i32> %tmp2, < i32 32, i32 32 >
@@ -105,6 +111,8 @@ define <2 x i32> @vsrau32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
}
define <1 x i64> @vsrau64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+;CHECK: vsrau64:
+;CHECK: vsra.u64
%tmp1 = load <1 x i64>* %A
%tmp2 = load <1 x i64>* %B
%tmp3 = lshr <1 x i64> %tmp2, < i64 64 >
@@ -113,6 +121,8 @@ define <1 x i64> @vsrau64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
}
define <16 x i8> @vsraQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+;CHECK: vsraQu8:
+;CHECK: vsra.u8
%tmp1 = load <16 x i8>* %A
%tmp2 = load <16 x i8>* %B
%tmp3 = lshr <16 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
@@ -121,6 +131,8 @@ define <16 x i8> @vsraQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
}
define <8 x i16> @vsraQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+;CHECK: vsraQu16:
+;CHECK: vsra.u16
%tmp1 = load <8 x i16>* %A
%tmp2 = load <8 x i16>* %B
%tmp3 = lshr <8 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16 >
@@ -129,6 +141,8 @@ define <8 x i16> @vsraQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
}
define <4 x i32> @vsraQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+;CHECK: vsraQu32:
+;CHECK: vsra.u32
%tmp1 = load <4 x i32>* %A
%tmp2 = load <4 x i32>* %B
%tmp3 = lshr <4 x i32> %tmp2, < i32 32, i32 32, i32 32, i32 32 >
@@ -137,6 +151,8 @@ define <4 x i32> @vsraQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
}
define <2 x i64> @vsraQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+;CHECK: vsraQu64:
+;CHECK: vsra.u64
%tmp1 = load <2 x i64>* %A
%tmp2 = load <2 x i64>* %B
%tmp3 = lshr <2 x i64> %tmp2, < i64 64, i64 64 >
@@ -145,6 +161,8 @@ define <2 x i64> @vsraQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
}
define <8 x i8> @vrsras8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+;CHECK: vrsras8:
+;CHECK: vrsra.s8
%tmp1 = load <8 x i8>* %A
%tmp2 = load <8 x i8>* %B
%tmp3 = call <8 x i8> @llvm.arm.neon.vrshifts.v8i8(<8 x i8> %tmp2, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
@@ -153,6 +171,8 @@ define <8 x i8> @vrsras8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
}
define <4 x i16> @vrsras16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+;CHECK: vrsras16:
+;CHECK: vrsra.s16
%tmp1 = load <4 x i16>* %A
%tmp2 = load <4 x i16>* %B
%tmp3 = call <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16> %tmp2, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
@@ -161,6 +181,8 @@ define <4 x i16> @vrsras16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
}
define <2 x i32> @vrsras32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+;CHECK: vrsras32:
+;CHECK: vrsra.s32
%tmp1 = load <2 x i32>* %A
%tmp2 = load <2 x i32>* %B
%tmp3 = call <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32> %tmp2, <2 x i32> < i32 -32, i32 -32 >)
@@ -169,6 +191,8 @@ define <2 x i32> @vrsras32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
}
define <1 x i64> @vrsras64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+;CHECK: vrsras64:
+;CHECK: vrsra.s64
%tmp1 = load <1 x i64>* %A
%tmp2 = load <1 x i64>* %B
%tmp3 = call <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64> %tmp2, <1 x i64> < i64 -64 >)
@@ -177,6 +201,8 @@ define <1 x i64> @vrsras64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
}
define <8 x i8> @vrsrau8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+;CHECK: vrsrau8:
+;CHECK: vrsra.u8
%tmp1 = load <8 x i8>* %A
%tmp2 = load <8 x i8>* %B
%tmp3 = call <8 x i8> @llvm.arm.neon.vrshiftu.v8i8(<8 x i8> %tmp2, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
@@ -185,6 +211,8 @@ define <8 x i8> @vrsrau8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
}
define <4 x i16> @vrsrau16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+;CHECK: vrsrau16:
+;CHECK: vrsra.u16
%tmp1 = load <4 x i16>* %A
%tmp2 = load <4 x i16>* %B
%tmp3 = call <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16> %tmp2, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
@@ -193,6 +221,8 @@ define <4 x i16> @vrsrau16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
}
define <2 x i32> @vrsrau32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+;CHECK: vrsrau32:
+;CHECK: vrsra.u32
%tmp1 = load <2 x i32>* %A
%tmp2 = load <2 x i32>* %B
%tmp3 = call <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32> %tmp2, <2 x i32> < i32 -32, i32 -32 >)
@@ -201,6 +231,8 @@ define <2 x i32> @vrsrau32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
}
define <1 x i64> @vrsrau64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+;CHECK: vrsrau64:
+;CHECK: vrsra.u64
%tmp1 = load <1 x i64>* %A
%tmp2 = load <1 x i64>* %B
%tmp3 = call <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64> %tmp2, <1 x i64> < i64 -64 >)
@@ -209,6 +241,8 @@ define <1 x i64> @vrsrau64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
}
define <16 x i8> @vrsraQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+;CHECK: vrsraQs8:
+;CHECK: vrsra.s8
%tmp1 = load <16 x i8>* %A
%tmp2 = load <16 x i8>* %B
%tmp3 = call <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8> %tmp2, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
@@ -217,6 +251,8 @@ define <16 x i8> @vrsraQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
}
define <8 x i16> @vrsraQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+;CHECK: vrsraQs16:
+;CHECK: vrsra.s16
%tmp1 = load <8 x i16>* %A
%tmp2 = load <8 x i16>* %B
%tmp3 = call <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16> %tmp2, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
@@ -225,6 +261,8 @@ define <8 x i16> @vrsraQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
}
define <4 x i32> @vrsraQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+;CHECK: vrsraQs32:
+;CHECK: vrsra.s32
%tmp1 = load <4 x i32>* %A
%tmp2 = load <4 x i32>* %B
%tmp3 = call <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32> %tmp2, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
@@ -233,6 +271,8 @@ define <4 x i32> @vrsraQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
}
define <2 x i64> @vrsraQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+;CHECK: vrsraQs64:
+;CHECK: vrsra.s64
%tmp1 = load <2 x i64>* %A
%tmp2 = load <2 x i64>* %B
%tmp3 = call <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64> %tmp2, <2 x i64> < i64 -64, i64 -64 >)
@@ -241,6 +281,8 @@ define <2 x i64> @vrsraQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
}
define <16 x i8> @vrsraQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+;CHECK: vrsraQu8:
+;CHECK: vrsra.u8
%tmp1 = load <16 x i8>* %A
%tmp2 = load <16 x i8>* %B
%tmp3 = call <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8> %tmp2, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
@@ -249,6 +291,8 @@ define <16 x i8> @vrsraQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
}
define <8 x i16> @vrsraQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+;CHECK: vrsraQu16:
+;CHECK: vrsra.u16
%tmp1 = load <8 x i16>* %A
%tmp2 = load <8 x i16>* %B
%tmp3 = call <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16> %tmp2, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
@@ -257,6 +301,8 @@ define <8 x i16> @vrsraQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
}
define <4 x i32> @vrsraQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+;CHECK: vrsraQu32:
+;CHECK: vrsra.u32
%tmp1 = load <4 x i32>* %A
%tmp2 = load <4 x i32>* %B
%tmp3 = call <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32> %tmp2, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
@@ -265,6 +311,8 @@ define <4 x i32> @vrsraQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
}
define <2 x i64> @vrsraQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+;CHECK: vrsraQu64:
+;CHECK: vrsra.u64
%tmp1 = load <2 x i64>* %A
%tmp2 = load <2 x i64>* %B
%tmp3 = call <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64> %tmp2, <2 x i64> < i64 -64, i64 -64 >)
OpenPOWER on IntegriCloud