summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorAndrea Di Biagio <Andrea_DiBiagio@sn.scee.net>2014-05-15 15:18:15 +0000
committerAndrea Di Biagio <Andrea_DiBiagio@sn.scee.net>2014-05-15 15:18:15 +0000
commit9836c47ea6184cadcbf8cc522e170960666935ee (patch)
treee25b351a75070abf8671d2afba8e6f479c6831b2
parent1f05322d96319651c484a90eead4cebd6bd67ad3 (diff)
[X86] Teach the backend how to fold SSE4.1/AVX/AVX2 blend intrinsics.
Added target specific combine rules to fold blend intrinsics according to the following rules: 1) fold(blend A, A, Mask) -> A; 2) fold(blend A, B, <allZeros>) -> A; 3) fold(blend A, B, <allOnes>) -> B. Added two new tests to verify that the new folding rules work for all the optimized blend intrinsics. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@208895 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/Target/X86/X86ISelLowering.cpp56
-rw-r--r--test/CodeGen/X86/combine-avx-intrinsics.ll119
-rw-r--r--test/CodeGen/X86/combine-avx2-intrinsics.ll113
-rw-r--r--test/CodeGen/X86/combine-sse41-intrinsics.ll182
4 files changed, 468 insertions, 2 deletions
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 4e9eecc15ed..8c9cc60f0f0 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -18473,10 +18473,61 @@ static SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG,
return SDValue();
}
-static SDValue PerformINTRINSIC_WO_CHAINCombine(SDNode *N, SelectionDAG &DAG) {
+static SDValue PerformINTRINSIC_WO_CHAINCombine(SDNode *N, SelectionDAG &DAG,
+ const X86Subtarget *Subtarget) {
unsigned IntNo = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
switch (IntNo) {
default: return SDValue();
+ // SSE/AVX/AVX2 blend intrinsics.
+ case Intrinsic::x86_avx2_pblendvb:
+ case Intrinsic::x86_avx2_pblendw:
+ case Intrinsic::x86_avx2_pblendd_128:
+ case Intrinsic::x86_avx2_pblendd_256:
+ // Don't try to simplify this intrinsic if we don't have AVX2.
+ if (!Subtarget->hasAVX2())
+ return SDValue();
+ // FALL-THROUGH
+ case Intrinsic::x86_avx_blend_pd_256:
+ case Intrinsic::x86_avx_blend_ps_256:
+ case Intrinsic::x86_avx_blendv_pd_256:
+ case Intrinsic::x86_avx_blendv_ps_256:
+ // Don't try to simplify this intrinsic if we don't have AVX.
+ if (!Subtarget->hasAVX())
+ return SDValue();
+ // FALL-THROUGH
+ case Intrinsic::x86_sse41_pblendw:
+ case Intrinsic::x86_sse41_blendpd:
+ case Intrinsic::x86_sse41_blendps:
+ case Intrinsic::x86_sse41_blendvps:
+ case Intrinsic::x86_sse41_blendvpd:
+ case Intrinsic::x86_sse41_pblendvb: {
+ SDValue Op0 = N->getOperand(1);
+ SDValue Op1 = N->getOperand(2);
+ SDValue Mask = N->getOperand(3);
+
+ // Don't try to simplify this intrinsic if we don't have SSE4.1.
+ if (!Subtarget->hasSSE41())
+ return SDValue();
+
+ // fold (blend A, A, Mask) -> A
+ if (Op0 == Op1)
+ return Op0;
+ // fold (blend A, B, allZeros) -> A
+ if (ISD::isBuildVectorAllZeros(Mask.getNode()))
+ return Op0;
+ // fold (blend A, B, allOnes) -> B
+ if (ISD::isBuildVectorAllOnes(Mask.getNode()))
+ return Op1;
+
+ // Simplify the case where the mask is a constant i32 value.
+ if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Mask)) {
+ if (C->isNullValue())
+ return Op0;
+ if (C->isAllOnesValue())
+ return Op1;
+ }
+ }
+
// Packed SSE2/AVX2 arithmetic shift immediate intrinsics.
case Intrinsic::x86_sse2_psrai_w:
case Intrinsic::x86_sse2_psrai_d:
@@ -20343,7 +20394,8 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
case X86ISD::VPERM2X128:
case ISD::VECTOR_SHUFFLE: return PerformShuffleCombine(N, DAG, DCI,Subtarget);
case ISD::FMA: return PerformFMACombine(N, DAG, Subtarget);
- case ISD::INTRINSIC_WO_CHAIN: return PerformINTRINSIC_WO_CHAINCombine(N, DAG);
+ case ISD::INTRINSIC_WO_CHAIN:
+ return PerformINTRINSIC_WO_CHAINCombine(N, DAG, Subtarget);
}
return SDValue();
diff --git a/test/CodeGen/X86/combine-avx-intrinsics.ll b/test/CodeGen/X86/combine-avx-intrinsics.ll
new file mode 100644
index 00000000000..f610f7fcb91
--- /dev/null
+++ b/test/CodeGen/X86/combine-avx-intrinsics.ll
@@ -0,0 +1,119 @@
+; RUN: llc < %s -march=x86-64 -mcpu=corei7-avx | FileCheck %s
+
+
+define <4 x double> @test_x86_avx_blend_pd_256(<4 x double> %a0) {
+ %1 = call <4 x double> @llvm.x86.avx.blend.pd.256(<4 x double> %a0, <4 x double> %a0, i32 7)
+ ret <4 x double> %1
+}
+; CHECK-LABEL: test_x86_avx_blend_pd_256
+; CHECK-NOT: vblendpd
+; CHECK: ret
+
+
+define <8 x float> @test_x86_avx_blend_ps_256(<8 x float> %a0) {
+ %1 = call <8 x float> @llvm.x86.avx.blend.ps.256(<8 x float> %a0, <8 x float> %a0, i32 7)
+ ret <8 x float> %1
+}
+; CHECK-LABEL: test_x86_avx_blend_ps_256
+; CHECK-NOT: vblendps
+; CHECK: ret
+
+
+define <4 x double> @test_x86_avx_blendv_pd_256(<4 x double> %a0, <4 x double> %a1) {
+ %1 = call <4 x double> @llvm.x86.avx.blendv.pd.256(<4 x double> %a0, <4 x double> %a0, <4 x double> %a1)
+ ret <4 x double> %1
+}
+; CHECK-LABEL: test_x86_avx_blendv_pd_256
+; CHECK-NOT: vblendvpd
+; CHECK: ret
+
+
+define <8 x float> @test_x86_avx_blendv_ps_256(<8 x float> %a0, <8 x float> %a1) {
+ %1 = call <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float> %a0, <8 x float> %a0, <8 x float> %a1)
+ ret <8 x float> %1
+}
+; CHECK-LABEL: test_x86_avx_blendv_ps_256
+; CHECK-NOT: vblendvps
+; CHECK: ret
+
+
+define <4 x double> @test2_x86_avx_blend_pd_256(<4 x double> %a0, <4 x double> %a1) {
+ %1 = call <4 x double> @llvm.x86.avx.blend.pd.256(<4 x double> %a0, <4 x double> %a1, i32 0)
+ ret <4 x double> %1
+}
+; CHECK-LABEL: test2_x86_avx_blend_pd_256
+; CHECK-NOT: vblendpd
+; CHECK: ret
+
+
+define <8 x float> @test2_x86_avx_blend_ps_256(<8 x float> %a0, <8 x float> %a1) {
+ %1 = call <8 x float> @llvm.x86.avx.blend.ps.256(<8 x float> %a0, <8 x float> %a1, i32 0)
+ ret <8 x float> %1
+}
+; CHECK-LABEL: test2_x86_avx_blend_ps_256
+; CHECK-NOT: vblendps
+; CHECK: ret
+
+
+define <4 x double> @test2_x86_avx_blendv_pd_256(<4 x double> %a0, <4 x double> %a1) {
+ %1 = call <4 x double> @llvm.x86.avx.blendv.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> zeroinitializer)
+ ret <4 x double> %1
+}
+; CHECK-LABEL: test2_x86_avx_blendv_pd_256
+; CHECK-NOT: vblendvpd
+; CHECK: ret
+
+
+define <8 x float> @test2_x86_avx_blendv_ps_256(<8 x float> %a0, <8 x float> %a1) {
+ %1 = call <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> zeroinitializer)
+ ret <8 x float> %1
+}
+; CHECK-LABEL: test2_x86_avx_blendv_ps_256
+; CHECK-NOT: vblendvps
+; CHECK: ret
+
+
+define <4 x double> @test3_x86_avx_blend_pd_256(<4 x double> %a0, <4 x double> %a1) {
+ %1 = call <4 x double> @llvm.x86.avx.blend.pd.256(<4 x double> %a0, <4 x double> %a1, i32 -1)
+ ret <4 x double> %1
+}
+; CHECK-LABEL: test3_x86_avx_blend_pd_256
+; CHECK-NOT: vblendpd
+; CHECK: ret
+
+
+define <8 x float> @test3_x86_avx_blend_ps_256(<8 x float> %a0, <8 x float> %a1) {
+ %1 = call <8 x float> @llvm.x86.avx.blend.ps.256(<8 x float> %a0, <8 x float> %a1, i32 -1)
+ ret <8 x float> %1
+}
+; CHECK-LABEL: test3_x86_avx_blend_ps_256
+; CHECK-NOT: vblendps
+; CHECK: ret
+
+
+define <4 x double> @test3_x86_avx_blendv_pd_256(<4 x double> %a0, <4 x double> %a1) {
+ %Mask = bitcast <4 x i64> <i64 -1, i64 -1, i64 -1, i64 -1> to <4 x double>
+ %1 = call <4 x double> @llvm.x86.avx.blendv.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %Mask)
+ ret <4 x double> %1
+}
+; CHECK-LABEL: test3_x86_avx_blendv_pd_256
+; CHECK-NOT: vblendvpd
+; CHECK: ret
+
+
+define <8 x float> @test3_x86_avx_blendv_ps_256(<8 x float> %a0, <8 x float> %a1) {
+ %Mask = bitcast <4 x i64> <i64 -1, i64 -1, i64 -1, i64 -1> to <8 x float>
+ %1 = call <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %Mask)
+ ret <8 x float> %1
+}
+; CHECK-LABEL: test3_x86_avx_blendv_ps_256
+; CHECK-NOT: vblendvps
+; CHECK: ret
+
+
+
+declare <4 x double> @llvm.x86.avx.blend.pd.256(<4 x double>, <4 x double>, i32)
+declare <8 x float> @llvm.x86.avx.blend.ps.256(<8 x float>, <8 x float>, i32)
+declare <4 x double> @llvm.x86.avx.blendv.pd.256(<4 x double>, <4 x double>, <4 x double>)
+declare <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float>, <8 x float>, <8 x float>)
+
diff --git a/test/CodeGen/X86/combine-avx2-intrinsics.ll b/test/CodeGen/X86/combine-avx2-intrinsics.ll
index 0560a8d6ae2..8794f8b8684 100644
--- a/test/CodeGen/X86/combine-avx2-intrinsics.ll
+++ b/test/CodeGen/X86/combine-avx2-intrinsics.ll
@@ -44,6 +44,119 @@ define <8 x i32> @test_psra_4(<8 x i32> %A) {
; CHECK: ret
+define <32 x i8> @test_x86_avx2_pblendvb(<32 x i8> %a0, <32 x i8> %a1) {
+ %res = call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a0, <32 x i8> %a0, <32 x i8> %a1)
+ ret <32 x i8> %res
+}
+; CHECK-LABEL: test_x86_avx2_pblendvb
+; CHECK-NOT: vpblendvb
+; CHECK: ret
+
+
+define <16 x i16> @test_x86_avx2_pblendw(<16 x i16> %a0) {
+ %res = call <16 x i16> @llvm.x86.avx2.pblendw(<16 x i16> %a0, <16 x i16> %a0, i32 7)
+ ret <16 x i16> %res
+}
+; CHECK-LABEL: test_x86_avx2_pblendw
+; CHECK-NOT: vpblendw
+; CHECK: ret
+
+
+define <4 x i32> @test_x86_avx2_pblendd_128(<4 x i32> %a0) {
+ %res = call <4 x i32> @llvm.x86.avx2.pblendd.128(<4 x i32> %a0, <4 x i32> %a0, i32 7)
+ ret <4 x i32> %res
+}
+; CHECK-LABEL: test_x86_avx2_pblendd_128
+; CHECK-NOT: vpblendd
+; CHECK: ret
+
+
+define <8 x i32> @test_x86_avx2_pblendd_256(<8 x i32> %a0) {
+ %res = call <8 x i32> @llvm.x86.avx2.pblendd.256(<8 x i32> %a0, <8 x i32> %a0, i32 7)
+ ret <8 x i32> %res
+}
+; CHECK-LABEL: test_x86_avx2_pblendd_256
+; CHECK-NOT: vpblendd
+; CHECK: ret
+
+
+define <32 x i8> @test2_x86_avx2_pblendvb(<32 x i8> %a0, <32 x i8> %a1) {
+ %res = call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> zeroinitializer)
+ ret <32 x i8> %res
+}
+; CHECK-LABEL: test2_x86_avx2_pblendvb
+; CHECK-NOT: vpblendvb
+; CHECK: ret
+
+
+define <16 x i16> @test2_x86_avx2_pblendw(<16 x i16> %a0, <16 x i16> %a1) {
+ %res = call <16 x i16> @llvm.x86.avx2.pblendw(<16 x i16> %a0, <16 x i16> %a1, i32 0)
+ ret <16 x i16> %res
+}
+; CHECK-LABEL: test2_x86_avx2_pblendw
+; CHECK-NOT: vpblendw
+; CHECK: ret
+
+
+define <4 x i32> @test2_x86_avx2_pblendd_128(<4 x i32> %a0, <4 x i32> %a1) {
+ %res = call <4 x i32> @llvm.x86.avx2.pblendd.128(<4 x i32> %a0, <4 x i32> %a1, i32 0)
+ ret <4 x i32> %res
+}
+; CHECK-LABEL: test2_x86_avx2_pblendd_128
+; CHECK-NOT: vpblendd
+; CHECK: ret
+
+
+define <8 x i32> @test2_x86_avx2_pblendd_256(<8 x i32> %a0, <8 x i32> %a1) {
+ %res = call <8 x i32> @llvm.x86.avx2.pblendd.256(<8 x i32> %a0, <8 x i32> %a1, i32 0)
+ ret <8 x i32> %res
+}
+; CHECK-LABEL: test2_x86_avx2_pblendd_256
+; CHECK-NOT: vpblendd
+; CHECK: ret
+
+
+define <32 x i8> @test3_x86_avx2_pblendvb(<32 x i8> %a0, <32 x i8> %a1) {
+ %1 = bitcast <4 x i64> <i64 -1, i64 -1, i64 -1, i64 -1> to <32 x i8>
+ %res = call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> %1)
+ ret <32 x i8> %res
+}
+; CHECK-LABEL: test3_x86_avx2_pblendvb
+; CHECK-NOT: vpblendvb
+; CHECK: ret
+
+
+define <16 x i16> @test3_x86_avx2_pblendw(<16 x i16> %a0, <16 x i16> %a1) {
+ %res = call <16 x i16> @llvm.x86.avx2.pblendw(<16 x i16> %a0, <16 x i16> %a1, i32 -1)
+ ret <16 x i16> %res
+}
+; CHECK-LABEL: test3_x86_avx2_pblendw
+; CHECK-NOT: vpblendw
+; CHECK: ret
+
+
+define <4 x i32> @test3_x86_avx2_pblendd_128(<4 x i32> %a0, <4 x i32> %a1) {
+ %res = call <4 x i32> @llvm.x86.avx2.pblendd.128(<4 x i32> %a0, <4 x i32> %a1, i32 -1)
+ ret <4 x i32> %res
+}
+; CHECK-LABEL: test3_x86_avx2_pblendd_128
+; CHECK-NOT: vpblendd
+; CHECK: ret
+
+
+define <8 x i32> @test3_x86_avx2_pblendd_256(<8 x i32> %a0, <8 x i32> %a1) {
+ %res = call <8 x i32> @llvm.x86.avx2.pblendd.256(<8 x i32> %a0, <8 x i32> %a1, i32 -1)
+ ret <8 x i32> %res
+}
+; CHECK-LABEL: test3_x86_avx2_pblendd_256
+; CHECK-NOT: vpblendd
+; CHECK: ret
+
+
+declare <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8>, <32 x i8>, <32 x i8>)
+declare <16 x i16> @llvm.x86.avx2.pblendw(<16 x i16>, <16 x i16>, i32)
+declare <4 x i32> @llvm.x86.avx2.pblendd.128(<4 x i32>, <4 x i32>, i32)
+declare <8 x i32> @llvm.x86.avx2.pblendd.256(<8 x i32>, <8 x i32>, i32)
declare <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16>, <8 x i16>)
declare <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16>, i32)
declare <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32>, <4 x i32>)
diff --git a/test/CodeGen/X86/combine-sse41-intrinsics.ll b/test/CodeGen/X86/combine-sse41-intrinsics.ll
new file mode 100644
index 00000000000..5dd5246c373
--- /dev/null
+++ b/test/CodeGen/X86/combine-sse41-intrinsics.ll
@@ -0,0 +1,182 @@
+; RUN: llc < %s -march=x86-64 -mcpu=corei7 | FileCheck %s
+
+
+define <2 x double> @test_x86_sse41_blend_pd(<2 x double> %a0, <2 x double> %a1) {
+ %1 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %a0, <2 x double> %a1, i32 0)
+ ret <2 x double> %1
+}
+; CHECK-LABEL: test_x86_sse41_blend_pd
+; CHECK-NOT: blendpd
+; CHECK: ret
+
+
+define <4 x float> @test_x86_sse41_blend_ps(<4 x float> %a0, <4 x float> %a1) {
+ %1 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %a0, <4 x float> %a1, i32 0)
+ ret <4 x float> %1
+}
+; CHECK-LABEL: test_x86_sse41_blend_ps
+; CHECK-NOT: blendps
+; CHECK: ret
+
+
+define <2 x double> @test_x86_sse41_blendv_pd(<2 x double> %a0, <2 x double> %a1) {
+ %1 = call <2 x double> @llvm.x86.sse41.blendvpd(<2 x double> %a0, <2 x double> %a1, <2 x double> zeroinitializer)
+ ret <2 x double> %1
+}
+; CHECK-LABEL: test_x86_sse41_blendv_pd
+; CHECK-NOT: blendvpd
+; CHECK: ret
+
+
+define <4 x float> @test_x86_sse41_blendv_ps(<4 x float> %a0, <4 x float> %a1) {
+ %1 = call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %a0, <4 x float> %a1, <4 x float> zeroinitializer)
+ ret <4 x float> %1
+}
+; CHECK-LABEL: test_x86_sse41_blendv_ps
+; CHECK-NOT: blendvps
+; CHECK: ret
+
+
+define <16 x i8> @test_x86_sse41_pblendv_b(<16 x i8> %a0, <16 x i8> %a1) {
+ %1 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> zeroinitializer)
+ ret <16 x i8> %1
+}
+; CHECK-LABEL: test_x86_sse41_pblendv_b
+; CHECK-NOT: pblendvb
+; CHECK: ret
+
+
+define <8 x i16> @test_x86_sse41_pblend_w(<8 x i16> %a0, <8 x i16> %a1) {
+ %1 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %a0, <8 x i16> %a1, i32 0)
+ ret <8 x i16> %1
+}
+; CHECK-LABEL: test_x86_sse41_pblend_w
+; CHECK-NOT: pblendw
+; CHECK: ret
+
+
+define <2 x double> @test2_x86_sse41_blend_pd(<2 x double> %a0, <2 x double> %a1) {
+ %1 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %a0, <2 x double> %a1, i32 -1)
+ ret <2 x double> %1
+}
+; CHECK-LABEL: test2_x86_sse41_blend_pd
+; CHECK-NOT: blendpd
+; CHECK: movaps %xmm1, %xmm0
+; CHECK-NEXT: ret
+
+
+define <4 x float> @test2_x86_sse41_blend_ps(<4 x float> %a0, <4 x float> %a1) {
+ %1 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %a0, <4 x float> %a1, i32 -1)
+ ret <4 x float> %1
+}
+; CHECK-LABEL: test2_x86_sse41_blend_ps
+; CHECK-NOT: blendps
+; CHECK: movaps %xmm1, %xmm0
+; CHECK-NEXT: ret
+
+
+define <2 x double> @test2_x86_sse41_blendv_pd(<2 x double> %a0, <2 x double> %a1) {
+ %Mask = bitcast <2 x i64> <i64 -1, i64 -1> to <2 x double>
+ %1 = call <2 x double> @llvm.x86.sse41.blendvpd(<2 x double> %a0, <2 x double> %a1, <2 x double> %Mask )
+ ret <2 x double> %1
+}
+; CHECK-LABEL: test2_x86_sse41_blendv_pd
+; CHECK-NOT: blendvpd
+; CHECK: movaps %xmm1, %xmm0
+; CHECK-NEXT: ret
+
+
+define <4 x float> @test2_x86_sse41_blendv_ps(<4 x float> %a0, <4 x float> %a1) {
+ %Mask = bitcast <2 x i64> <i64 -1, i64 -1> to <4 x float>
+ %1 = call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %a0, <4 x float> %a1, <4 x float> %Mask)
+ ret <4 x float> %1
+}
+; CHECK-LABEL: test2_x86_sse41_blendv_ps
+; CHECK-NOT: blendvps
+; CHECK: movaps %xmm1, %xmm0
+; CHECK-NEXT: ret
+
+
+define <16 x i8> @test2_x86_sse41_pblendv_b(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %a2) {
+ %Mask = bitcast <2 x i64> <i64 -1, i64 -1> to <16 x i8>
+ %1 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %Mask)
+ ret <16 x i8> %1
+}
+; CHECK-LABEL: test2_x86_sse41_pblendv_b
+; CHECK-NOT: pblendvb
+; CHECK: movaps %xmm1, %xmm0
+; CHECK-NEXT: ret
+
+
+define <8 x i16> @test2_x86_sse41_pblend_w(<8 x i16> %a0, <8 x i16> %a1) {
+ %1 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %a0, <8 x i16> %a1, i32 -1)
+ ret <8 x i16> %1
+}
+; CHECK-LABEL: test2_x86_sse41_pblend_w
+; CHECK-NOT: pblendw
+; CHECK: movaps %xmm1, %xmm0
+; CHECK-NEXT: ret
+
+
+define <2 x double> @test3_x86_sse41_blend_pd(<2 x double> %a0) {
+ %1 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %a0, <2 x double> %a0, i32 7)
+ ret <2 x double> %1
+}
+; CHECK-LABEL: test3_x86_sse41_blend_pd
+; CHECK-NOT: blendpd
+; CHECK: ret
+
+
+define <4 x float> @test3_x86_sse41_blend_ps(<4 x float> %a0) {
+ %1 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %a0, <4 x float> %a0, i32 7)
+ ret <4 x float> %1
+}
+; CHECK-LABEL: test3_x86_sse41_blend_ps
+; CHECK-NOT: blendps
+; CHECK: ret
+
+
+define <2 x double> @test3_x86_sse41_blendv_pd(<2 x double> %a0, <2 x double> %a1) {
+ %1 = call <2 x double> @llvm.x86.sse41.blendvpd(<2 x double> %a0, <2 x double> %a0, <2 x double> %a1 )
+ ret <2 x double> %1
+}
+; CHECK-LABEL: test3_x86_sse41_blendv_pd
+; CHECK-NOT: blendvpd
+; CHECK: ret
+
+
+define <4 x float> @test3_x86_sse41_blendv_ps(<4 x float> %a0, <4 x float> %a1) {
+ %1 = call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %a0, <4 x float> %a0, <4 x float> %a1)
+ ret <4 x float> %1
+}
+; CHECK-LABEL: test3_x86_sse41_blendv_ps
+; CHECK-NOT: blendvps
+; CHECK: ret
+
+
+define <16 x i8> @test3_x86_sse41_pblendv_b(<16 x i8> %a0, <16 x i8> %a1) {
+ %1 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a0, <16 x i8> %a0, <16 x i8> %a1)
+ ret <16 x i8> %1
+}
+; CHECK-LABEL: test3_x86_sse41_pblendv_b
+; CHECK-NOT: pblendvb
+; CHECK: ret
+
+
+define <8 x i16> @test3_x86_sse41_pblend_w(<8 x i16> %a0) {
+ %1 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %a0, <8 x i16> %a0, i32 7)
+ ret <8 x i16> %1
+}
+; CHECK-LABEL: test3_x86_sse41_pblend_w
+; CHECK-NOT: pblendw
+; CHECK: ret
+
+
+declare <2 x double> @llvm.x86.sse41.blendpd(<2 x double>, <2 x double>, i32)
+declare <4 x float> @llvm.x86.sse41.blendps(<4 x float>, <4 x float>, i32)
+declare <2 x double> @llvm.x86.sse41.blendvpd(<2 x double>, <2 x double>, <2 x double>)
+declare <4 x float> @llvm.x86.sse41.blendvps(<4 x float>, <4 x float>, <4 x float>)
+declare <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8>, <16 x i8>, <16 x i8>)
+declare <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16>, <8 x i16>, i32)
+declare <8 x i16> @llvm.x86.sse41.phminposuw(<8 x i16>)
+