RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-31 12:41:49 +01:00
Code Issues Projects Releases Wiki Activity

AMDGPU: Custom lower vector_shuffle for v4i16/v4f16

Browse Source 
Ordinarily it is lowered as a build_vector of each extract_vector_elt,
which in turn get lowered to bitcasts and bit shifts. Very little
understand the lowered extract pattern, resulting in much worse
code. We treat concat_vectors of v2i16 as legal, so prefer that.

llvm-svn: 364959
This commit is contained in:
Matt Arsenault 2019-07-02 19:15:45 +00:00
parent a34bcd1ee4
commit 0b429caefd
3 changed files with 146 additions and 189 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
lib/Target/AMDGPU/SIISelLowering.cpp
View File

@ -630,6 +630,9 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2i16, Custom);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2f16, Custom);
setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v4f16, Custom);
setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v4i16, Custom);
setOperationAction(ISD::SHL, MVT::v4i16, Custom);
setOperationAction(ISD::SRA, MVT::v4i16, Custom);
setOperationAction(ISD::SRL, MVT::v4i16, Custom);
@ -3957,6 +3960,8 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
return lowerINSERT_VECTOR_ELT(Op, DAG);
case ISD::EXTRACT_VECTOR_ELT:
return lowerEXTRACT_VECTOR_ELT(Op, DAG);
case ISD::VECTOR_SHUFFLE:
return lowerVECTOR_SHUFFLE(Op, DAG);
case ISD::BUILD_VECTOR:
return lowerBUILD_VECTOR(Op, DAG);
case ISD::FP_ROUND:
@ -4740,6 +4745,63 @@ SDValue SITargetLowering::lowerEXTRACT_VECTOR_ELT(SDValue Op,
return DAG.getAnyExtOrTrunc(Elt, SL, ResultVT);
}
static bool elementPairIsContiguous(ArrayRef<int> Mask, int Elt) {
assert(Elt % 2 == 0);
return Mask[Elt + 1] == Mask[Elt] + 1 && (Mask[Elt] % 2 == 0);
}
SDValue SITargetLowering::lowerVECTOR_SHUFFLE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc SL(Op);
EVT ResultVT = Op.getValueType();
ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op);
EVT PackVT = ResultVT.isInteger() ? MVT::v2i16 : MVT::v2f16;
EVT EltVT = PackVT.getVectorElementType();
int SrcNumElts = Op.getOperand(0).getValueType().getVectorNumElements();
// vector_shuffle <0,1,6,7> lhs, rhs
// -> concat_vectors (extract_subvector lhs, 0), (extract_subvector rhs, 2)
//
// vector_shuffle <6,7,2,3> lhs, rhs
// -> concat_vectors (extract_subvector rhs, 2), (extract_subvector lhs, 2)
//
// vector_shuffle <6,7,0,1> lhs, rhs
// -> concat_vectors (extract_subvector rhs, 2), (extract_subvector lhs, 0)
// Avoid scalarizing when both halves are reading from consecutive elements.
SmallVector<SDValue, 4> Pieces;
for (int I = 0, N = ResultVT.getVectorNumElements(); I != N; I += 2) {
if (elementPairIsContiguous(SVN->getMask(), I)) {
const int Idx = SVN->getMaskElt(I);
int VecIdx = Idx < SrcNumElts ? 0 : 1;
int EltIdx = Idx < SrcNumElts ? Idx : Idx - SrcNumElts;
SDValue SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, SL,
PackVT, SVN->getOperand(VecIdx),
DAG.getConstant(EltIdx, SL, MVT::i32));
Pieces.push_back(SubVec);
} else {
const int Idx0 = SVN->getMaskElt(I);
const int Idx1 = SVN->getMaskElt(I + 1);
int VecIdx0 = Idx0 < SrcNumElts ? 0 : 1;
int VecIdx1 = Idx1 < SrcNumElts ? 0 : 1;
int EltIdx0 = Idx0 < SrcNumElts ? Idx0 : Idx0 - SrcNumElts;
int EltIdx1 = Idx1 < SrcNumElts ? Idx1 : Idx1 - SrcNumElts;
SDValue Vec0 = SVN->getOperand(VecIdx0);
SDValue Elt0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, EltVT,
Vec0, DAG.getConstant(EltIdx0, SL, MVT::i32));
SDValue Vec1 = SVN->getOperand(VecIdx1);
SDValue Elt1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, EltVT,
Vec1, DAG.getConstant(EltIdx1, SL, MVT::i32));
Pieces.push_back(DAG.getBuildVector(PackVT, SL, { Elt0, Elt1 }));
}
}
return DAG.getNode(ISD::CONCAT_VECTORS, SL, ResultVT, Pieces);
}
SDValue SITargetLowering::lowerBUILD_VECTOR(SDValue Op,
SelectionDAG &DAG) const {
SDLoc SL(Op);

1
lib/Target/AMDGPU/SIISelLowering.h
View File

@ -123,6 +123,7 @@ private:
SDValue lowerADDRSPACECAST(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerTRAP(SDValue Op, SelectionDAG &DAG) const;
SDValue lowerDEBUGTRAP(SDValue Op, SelectionDAG &DAG) const;

272
test/CodeGen/AMDGPU/vector_shuffle.packed.ll
View File

@ -7,9 +7,7 @@ define <4 x half> @shuffle_v4f16_23uu(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v1
; GFX9-NEXT: v_lshl_or_b32 v0, v0, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -24,10 +22,8 @@ define <4 x half> @shuffle_v4f16_234u(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: global_load_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v5
; GFX9-NEXT: v_mov_b32_e32 v0, v5
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_e32 v2, 0xffff, v5
; GFX9-NEXT: v_lshl_or_b32 v0, v0, 16, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -160,9 +156,6 @@ define <4 x half> @shuffle_v4f16_0101(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v0
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@ -192,13 +185,6 @@ define <4 x half> @shuffle_v4f16_0145(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, v2, v0
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v1
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -214,13 +200,7 @@ define <4 x half> @shuffle_v4f16_0167(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, v1, v0
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2
; GFX9-NEXT: v_and_b32_e32 v1, v1, v2
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v1, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -232,15 +212,9 @@ define <4 x half> @shuffle_v4f16_2301(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_2301:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v0
; GFX9-NEXT: v_and_b32_e32 v2, v2, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v2
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -254,10 +228,7 @@ define <4 x half> @shuffle_v4f16_2323(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v1
; GFX9-NEXT: v_lshl_or_b32 v0, v0, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: v_mov_b32_e32 v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -269,17 +240,11 @@ define <4 x half> @shuffle_v4f16_2345(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_2345:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[2:3], v[2:3], off
; GFX9-NEXT: global_load_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: v_mov_b32_e32 v0, v5
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2
; GFX9-NEXT: v_and_b32_e32 v1, v0, v1
; GFX9-NEXT: v_and_b32_e32 v2, v0, v2
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -291,17 +256,10 @@ define <4 x half> @shuffle_v4f16_2367(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_2367:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[2:3], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: global_load_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v3
; GFX9-NEXT: v_and_b32_e32 v1, v0, v1
; GFX9-NEXT: v_and_b32_e32 v3, v0, v3
; GFX9-NEXT: v_lshl_or_b32 v0, v2, 16, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v3
; GFX9-NEXT: v_mov_b32_e32 v0, v5
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -313,17 +271,10 @@ define <4 x half> @shuffle_v4f16_4501(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_4501:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v1
; GFX9-NEXT: v_and_b32_e32 v2, v2, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v4, 16, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v2
; GFX9-NEXT: v_mov_b32_e32 v1, v4
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -337,15 +288,8 @@ define <4 x half> @shuffle_v4f16_4523(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[2:3], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v0, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2
; GFX9-NEXT: v_and_b32_e32 v2, v0, v2
; GFX9-NEXT: v_lshl_or_b32 v0, v4, 16, v2
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -359,9 +303,6 @@ define <4 x half> @shuffle_v4f16_4545(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v0
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@ -387,17 +328,11 @@ define <4 x half> @shuffle_v4f16_6701(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_6701:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2
; GFX9-NEXT: v_and_b32_e32 v2, v1, v2
; GFX9-NEXT: v_and_b32_e32 v1, v1, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v4, 16, v2
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v0, v1
; GFX9-NEXT: v_mov_b32_e32 v1, v4
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -411,15 +346,8 @@ define <4 x half> @shuffle_v4f16_6723(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[2:3], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v0, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v3
; GFX9-NEXT: v_and_b32_e32 v3, v0, v3
; GFX9-NEXT: v_lshl_or_b32 v0, v4, 16, v3
; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v0, v3
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -431,15 +359,9 @@ define <4 x half> @shuffle_v4f16_6745(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_6745:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v0
; GFX9-NEXT: v_and_b32_e32 v2, v2, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v2
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -453,10 +375,7 @@ define <4 x half> @shuffle_v4f16_6767(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v1
; GFX9-NEXT: v_lshl_or_b32 v0, v0, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: v_mov_b32_e32 v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -468,16 +387,13 @@ define <4 x half> @shuffle_v4f16_2356(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_2356:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[2:3], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v0, v1
; GFX9-NEXT: global_load_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v0, v4, 16, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v2
; GFX9-NEXT: v_and_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v1, v1, 16, v0
; GFX9-NEXT: v_mov_b32_e32 v0, v5
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -493,12 +409,9 @@ define <4 x half> @shuffle_v4f16_5623(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-NEXT: global_load_dwordx2 v[2:3], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v0, v1
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v2
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v1
; GFX9-NEXT: v_and_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -571,14 +484,13 @@ define <4 x i16> @shuffle_v4i16_2356(<4 x i16> addrspace(1)* %arg0, <4 x i16> ad
; GFX9-LABEL: shuffle_v4i16_2356:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[2:3], v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v4, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_bfi_b32 v0, v4, v1, v1
; GFX9-NEXT: global_load_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_sdwa v1, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v1
; GFX9-NEXT: v_and_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v1, v1, 16, v0
; GFX9-NEXT: v_mov_b32_e32 v0, v5
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x i16>, <4 x i16> addrspace(1)* %arg0
%val1 = load <4 x i16>, <4 x i16> addrspace(1)* %arg1
@ -594,9 +506,7 @@ define <4 x i16> @shuffle_v4i16_0167(<4 x i16> addrspace(1)* %arg0, <4 x i16> ad
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff
; GFX9-NEXT: v_bfi_b32 v0, v1, v0, v0
; GFX9-NEXT: v_bfi_b32 v1, v1, v2, v2
; GFX9-NEXT: v_mov_b32_e32 v1, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x i16>, <4 x i16> addrspace(1)* %arg0
%val1 = load <4 x i16>, <4 x i16> addrspace(1)* %arg1
@ -679,14 +589,12 @@ define <4 x half> @shuffle_v4f16_2333(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_2333:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-NEXT: v_and_b32_e32 v0, v2, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v3
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v2
; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v0, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -698,14 +606,12 @@ define <4 x half> @shuffle_v4f16_6667(<4 x half> addrspace(1)* %arg0, <4 x half>
; GFX9-LABEL: shuffle_v4f16_6667:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-NEXT: v_and_b32_e32 v0, v2, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v3
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v2
; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v0, 16, v1
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@ -719,9 +625,6 @@ define <4 x half> @shuffle_v8f16_0101(<8 x half> addrspace(1)* %arg0, <8 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dword v0, v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v0
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
@ -749,15 +652,7 @@ define <4 x half> @shuffle_v8f16_4589(<8 x half> addrspace(1)* %arg0, <8 x half>
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dword v0, v[0:1], off offset:8
; GFX9-NEXT: global_load_dword v1, v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, v2, v0
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v1
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
%val1 = load <8 x half>, <8 x half> addrspace(1)* %arg1
@ -769,17 +664,9 @@ define <4 x half> @shuffle_v8f16_10_11_2_3(<8 x half> addrspace(1)* %arg0, <8 x
; GFX9-LABEL: shuffle_v8f16_10_11_2_3:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dword v0, v[0:1], off offset:4
; GFX9-NEXT: global_load_dword v1, v[2:3], off offset:4
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX9-NEXT: global_load_dword v1, v[0:1], off offset:4
; GFX9-NEXT: global_load_dword v0, v[2:3], off offset:4
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v1
; GFX9-NEXT: v_and_b32_e32 v1, v2, v1
; GFX9-NEXT: v_and_b32_e32 v2, v2, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v4, 16, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v3, 16, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
%val1 = load <8 x half>, <8 x half> addrspace(1)* %arg1
@ -791,15 +678,12 @@ define <4 x half> @shuffle_v8f16_13_14_2_3(<8 x half> addrspace(1)* %arg0, <8 x
; GFX9-LABEL: shuffle_v8f16_13_14_2_3:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dword v4, v[0:1], off offset:4
; GFX9-NEXT: global_load_dwordx4 v[0:3], v[2:3], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_dword v1, v[0:1], off offset:4
; GFX9-NEXT: global_load_dwordx4 v[2:5], v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v4
; GFX9-NEXT: v_and_b32_e32 v4, v0, v4
; GFX9-NEXT: v_and_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v1, v1, 16, v4
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v0, v5, 16, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
%val1 = load <8 x half>, <8 x half> addrspace(1)* %arg1
@ -812,12 +696,8 @@ define <4 x half> @shuffle_v3f16_0122(<3 x half> addrspace(1)* %arg0, <3 x half>
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, v2, v0
; GFX9-NEXT: v_and_b32_e32 v2, v2, v1
; GFX9-NEXT: v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT: v_and_b32_e32 v2, 0xffff, v1
; GFX9-NEXT: v_lshl_or_b32 v1, v1, 16, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <3 x half>, <3 x half> addrspace(1)* %arg0
@ -846,17 +726,14 @@ define <6 x half> @shuffle_v6f16_452367(<6 x half> addrspace(1)* %arg0, <6 x hal
; GFX9-LABEL: shuffle_v6f16_452367:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx3 v[4:6], v[0:1], off
; GFX9-NEXT: global_load_dword v2, v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v0, 0xffff
; GFX9-NEXT: v_mov_b32_e32 v4, v3
; GFX9-NEXT: v_mov_b32_e32 v3, v2
; GFX9-NEXT: global_load_dwordx3 v[0:2], v[0:1], off
; GFX9-NEXT: global_load_dword v3, v[3:4], off
; GFX9-NEXT: s_waitcnt vmcnt(1)
; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v5
; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v6
; GFX9-NEXT: v_and_b32_e32 v3, v0, v6
; GFX9-NEXT: v_and_b32_e32 v5, v0, v5
; GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v3
; GFX9-NEXT: v_lshl_or_b32 v1, v4, 16, v5
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v2, v3
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <6 x half>, <6 x half> addrspace(1)* %arg0
%val1 = load <6 x half>, <6 x half> addrspace(1)* %arg1
@ -888,9 +765,6 @@ define amdgpu_kernel void @fma_shuffle(<4 x half> addrspace(1)* nocapture readon
; GFX9-NEXT: v_pk_fma_f16 v2, v1, v2, v7 op_sel_hi:[0,1,1]
; GFX9-NEXT: v_pk_fma_f16 v0, v0, v3, v6 op_sel:[1,0,0]
; GFX9-NEXT: v_pk_fma_f16 v1, v1, v3, v2 op_sel:[1,0,0]
; GFX9-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX9-NEXT: v_lshl_or_b32 v0, v2, 16, v0
; GFX9-NEXT: global_store_dwordx2 v[4:5], v[0:1], off
; GFX9-NEXT: s_endpgm
entry:
@ -922,6 +796,26 @@ entry:
ret void
}
define <4 x half> @shuffle_v4f16_0456(<4 x half> addrspace(1)* %arg0, <4 x half> addrspace(1)* %arg1) {
; GFX9-LABEL: shuffle_v4f16_0456:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_dwordx2 v[1:2], v[2:3], off
; GFX9-NEXT: v_mov_b32_e32 v3, 0xffff
; GFX9-NEXT: v_and_b32_e32 v0, v3, v0
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_sdwa v3, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v0
; GFX9-NEXT: v_lshl_or_b32 v1, v2, 16, v3
; GFX9-NEXT: s_setpc_b64 s[30:31]
%val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
%val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
%shuffle = shufflevector <4 x half> %val0, <4 x half> %val1, <4 x i32> <i32 0, i32 4, i32 5, i32 6>
ret <4 x half> %shuffle
}
declare <2 x half> @llvm.fma.v2f16(<2 x half>, <2 x half>, <2 x half>) #0
declare i32 @llvm.amdgcn.workitem.id.x() #0