2009-09-09 00:09:15 +00:00
|
|
|
; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
|
2009-08-06 18:47:44 +00:00
|
|
|
|
|
|
|
define void @vst2i8(i8* %A, <8 x i8>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2i8:
|
2010-09-23 23:42:37 +00:00
|
|
|
;Check the alignment value. Max for this instruction is 128 bits:
|
2013-02-22 10:01:33 +00:00
|
|
|
;CHECK: vst2.8 {d16, d17}, [r0:64]
|
2009-08-06 18:47:44 +00:00
|
|
|
%tmp1 = load <8 x i8>* %B
|
2010-09-23 23:42:37 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v8i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 8)
|
2009-08-06 18:47:44 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2011-02-07 17:43:21 +00:00
|
|
|
;Check for a post-increment updating store with register increment.
|
|
|
|
define void @vst2i8_update(i8** %ptr, <8 x i8>* %B, i32 %inc) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2i8_update:
|
2011-02-07 17:43:21 +00:00
|
|
|
;CHECK: vst2.8 {d16, d17}, [r1], r2
|
|
|
|
%A = load i8** %ptr
|
|
|
|
%tmp1 = load <8 x i8>* %B
|
|
|
|
call void @llvm.arm.neon.vst2.v8i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 4)
|
|
|
|
%tmp2 = getelementptr i8* %A, i32 %inc
|
|
|
|
store i8* %tmp2, i8** %ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2009-08-06 18:47:44 +00:00
|
|
|
define void @vst2i16(i16* %A, <4 x i16>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2i16:
|
2010-09-23 23:42:37 +00:00
|
|
|
;Check the alignment value. Max for this instruction is 128 bits:
|
2013-02-22 10:01:33 +00:00
|
|
|
;CHECK: vst2.16 {d16, d17}, [r0:128]
|
2010-04-20 00:17:16 +00:00
|
|
|
%tmp0 = bitcast i16* %A to i8*
|
2009-08-06 18:47:44 +00:00
|
|
|
%tmp1 = load <4 x i16>* %B
|
2010-09-23 23:42:37 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v4i16(i8* %tmp0, <4 x i16> %tmp1, <4 x i16> %tmp1, i32 32)
|
2009-08-06 18:47:44 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @vst2i32(i32* %A, <2 x i32>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2i32:
|
2009-08-06 18:47:44 +00:00
|
|
|
;CHECK: vst2.32
|
2010-04-20 00:17:16 +00:00
|
|
|
%tmp0 = bitcast i32* %A to i8*
|
2009-08-06 18:47:44 +00:00
|
|
|
%tmp1 = load <2 x i32>* %B
|
2010-08-27 17:13:24 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v2i32(i8* %tmp0, <2 x i32> %tmp1, <2 x i32> %tmp1, i32 1)
|
2009-08-06 18:47:44 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @vst2f(float* %A, <2 x float>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2f:
|
2009-08-06 18:47:44 +00:00
|
|
|
;CHECK: vst2.32
|
2010-04-20 00:17:16 +00:00
|
|
|
%tmp0 = bitcast float* %A to i8*
|
2009-08-06 18:47:44 +00:00
|
|
|
%tmp1 = load <2 x float>* %B
|
2010-08-27 17:13:24 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v2f32(i8* %tmp0, <2 x float> %tmp1, <2 x float> %tmp1, i32 1)
|
2009-08-06 18:47:44 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2009-10-08 00:21:01 +00:00
|
|
|
define void @vst2i64(i64* %A, <1 x i64>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2i64:
|
2010-09-23 23:42:37 +00:00
|
|
|
;Check the alignment value. Max for this instruction is 128 bits:
|
2013-02-22 10:01:33 +00:00
|
|
|
;CHECK: vst1.64 {d16, d17}, [r0:128]
|
2010-04-20 00:17:16 +00:00
|
|
|
%tmp0 = bitcast i64* %A to i8*
|
2009-10-08 00:21:01 +00:00
|
|
|
%tmp1 = load <1 x i64>* %B
|
2010-09-23 23:42:37 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v1i64(i8* %tmp0, <1 x i64> %tmp1, <1 x i64> %tmp1, i32 32)
|
2009-10-08 00:21:01 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2011-02-07 17:43:21 +00:00
|
|
|
;Check for a post-increment updating store.
|
|
|
|
define void @vst2i64_update(i64** %ptr, <1 x i64>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2i64_update:
|
2013-02-22 10:01:33 +00:00
|
|
|
;CHECK: vst1.64 {d16, d17}, [r1:64]!
|
2011-02-07 17:43:21 +00:00
|
|
|
%A = load i64** %ptr
|
|
|
|
%tmp0 = bitcast i64* %A to i8*
|
|
|
|
%tmp1 = load <1 x i64>* %B
|
|
|
|
call void @llvm.arm.neon.vst2.v1i64(i8* %tmp0, <1 x i64> %tmp1, <1 x i64> %tmp1, i32 8)
|
|
|
|
%tmp2 = getelementptr i64* %A, i32 2
|
|
|
|
store i64* %tmp2, i64** %ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2009-10-07 18:47:39 +00:00
|
|
|
define void @vst2Qi8(i8* %A, <16 x i8>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2Qi8:
|
2010-09-23 23:42:37 +00:00
|
|
|
;Check the alignment value. Max for this instruction is 256 bits:
|
2013-02-22 10:01:33 +00:00
|
|
|
;CHECK: vst2.8 {d16, d17, d18, d19}, [r0:64]
|
2009-10-07 18:47:39 +00:00
|
|
|
%tmp1 = load <16 x i8>* %B
|
2010-09-23 23:42:37 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v16i8(i8* %A, <16 x i8> %tmp1, <16 x i8> %tmp1, i32 8)
|
2009-10-07 18:47:39 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @vst2Qi16(i16* %A, <8 x i16>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2Qi16:
|
2010-09-23 23:42:37 +00:00
|
|
|
;Check the alignment value. Max for this instruction is 256 bits:
|
2013-02-22 10:01:33 +00:00
|
|
|
;CHECK: vst2.16 {d16, d17, d18, d19}, [r0:128]
|
2010-04-20 00:17:16 +00:00
|
|
|
%tmp0 = bitcast i16* %A to i8*
|
2009-10-07 18:47:39 +00:00
|
|
|
%tmp1 = load <8 x i16>* %B
|
2010-09-23 23:42:37 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v8i16(i8* %tmp0, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 16)
|
2009-10-07 18:47:39 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @vst2Qi32(i32* %A, <4 x i32>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2Qi32:
|
2010-09-23 23:42:37 +00:00
|
|
|
;Check the alignment value. Max for this instruction is 256 bits:
|
2013-02-22 10:01:33 +00:00
|
|
|
;CHECK: vst2.32 {d16, d17, d18, d19}, [r0:256]
|
2010-04-20 00:17:16 +00:00
|
|
|
%tmp0 = bitcast i32* %A to i8*
|
2009-10-07 18:47:39 +00:00
|
|
|
%tmp1 = load <4 x i32>* %B
|
2010-09-23 23:42:37 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v4i32(i8* %tmp0, <4 x i32> %tmp1, <4 x i32> %tmp1, i32 64)
|
2009-10-07 18:47:39 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @vst2Qf(float* %A, <4 x float>* %B) nounwind {
|
2013-07-14 06:24:09 +00:00
|
|
|
;CHECK-LABEL: vst2Qf:
|
2009-10-07 18:47:39 +00:00
|
|
|
;CHECK: vst2.32
|
2010-04-20 00:17:16 +00:00
|
|
|
%tmp0 = bitcast float* %A to i8*
|
2009-10-07 18:47:39 +00:00
|
|
|
%tmp1 = load <4 x float>* %B
|
2010-08-27 17:13:24 +00:00
|
|
|
call void @llvm.arm.neon.vst2.v4f32(i8* %tmp0, <4 x float> %tmp1, <4 x float> %tmp1, i32 1)
|
2009-10-07 18:47:39 +00:00
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2012-01-10 21:11:12 +00:00
|
|
|
define i8* @vst2update(i8* %out, <4 x i16>* %B) nounwind {
|
2013-07-18 22:47:09 +00:00
|
|
|
;CHECK-LABEL: vst2update:
|
2012-01-10 21:11:12 +00:00
|
|
|
;CHECK: vst2.16 {d16, d17}, [r0]!
|
|
|
|
%tmp1 = load <4 x i16>* %B
|
|
|
|
tail call void @llvm.arm.neon.vst2.v4i16(i8* %out, <4 x i16> %tmp1, <4 x i16> %tmp1, i32 2)
|
|
|
|
%t5 = getelementptr inbounds i8* %out, i32 16
|
|
|
|
ret i8* %t5
|
|
|
|
}
|
|
|
|
|
2012-01-20 19:16:00 +00:00
|
|
|
define i8* @vst2update2(i8 * %out, <4 x float> * %this) nounwind optsize ssp align 2 {
|
2013-07-18 22:47:09 +00:00
|
|
|
;CHECK-LABEL: vst2update2:
|
2012-01-20 19:16:00 +00:00
|
|
|
;CHECK: vst2.32 {d16, d17, d18, d19}, [r0]!
|
|
|
|
%tmp1 = load <4 x float>* %this
|
|
|
|
call void @llvm.arm.neon.vst2.v4f32(i8* %out, <4 x float> %tmp1, <4 x float> %tmp1, i32 4) nounwind
|
|
|
|
%tmp2 = getelementptr inbounds i8* %out, i32 32
|
|
|
|
ret i8* %tmp2
|
|
|
|
}
|
|
|
|
|
2010-08-27 17:13:24 +00:00
|
|
|
declare void @llvm.arm.neon.vst2.v8i8(i8*, <8 x i8>, <8 x i8>, i32) nounwind
|
|
|
|
declare void @llvm.arm.neon.vst2.v4i16(i8*, <4 x i16>, <4 x i16>, i32) nounwind
|
|
|
|
declare void @llvm.arm.neon.vst2.v2i32(i8*, <2 x i32>, <2 x i32>, i32) nounwind
|
|
|
|
declare void @llvm.arm.neon.vst2.v2f32(i8*, <2 x float>, <2 x float>, i32) nounwind
|
|
|
|
declare void @llvm.arm.neon.vst2.v1i64(i8*, <1 x i64>, <1 x i64>, i32) nounwind
|
2009-10-07 18:47:39 +00:00
|
|
|
|
2010-08-27 17:13:24 +00:00
|
|
|
declare void @llvm.arm.neon.vst2.v16i8(i8*, <16 x i8>, <16 x i8>, i32) nounwind
|
|
|
|
declare void @llvm.arm.neon.vst2.v8i16(i8*, <8 x i16>, <8 x i16>, i32) nounwind
|
|
|
|
declare void @llvm.arm.neon.vst2.v4i32(i8*, <4 x i32>, <4 x i32>, i32) nounwind
|
|
|
|
declare void @llvm.arm.neon.vst2.v4f32(i8*, <4 x float>, <4 x float>, i32) nounwind
|