[AMDGPU] Implemented dwordx3 variants of buffer/tbuffer load/store intrinsics

Now we have vec3 MVTs, this commit implements dwordx3 variants of the
buffer intrinsics.

On gfx6, a dwordx3 buffer load intrinsic is implemented as a dwordx4
instruction, and a dwordx3 buffer store intrinsic is not supported.
We need to support the dwordx3 load intrinsic because it is generated by
subtarget-unaware code in InstCombine.

Differential Revision: https://reviews.llvm.org/D58904

Change-Id: I016729d8557b98a52f529638ae97c340a5922a4e
llvm-svn: 356755

6 files changed, 68 insertions, 24 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 3d460199d94..623d9817f45 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -4295,7 +4295,6 @@ const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
   NODE_NAME_CASE(STORE_MSKOR)
   NODE_NAME_CASE(LOAD_CONSTANT)
   NODE_NAME_CASE(TBUFFER_STORE_FORMAT)
-  NODE_NAME_CASE(TBUFFER_STORE_FORMAT_X3)
   NODE_NAME_CASE(TBUFFER_STORE_FORMAT_D16)
   NODE_NAME_CASE(TBUFFER_LOAD_FORMAT)
   NODE_NAME_CASE(TBUFFER_LOAD_FORMAT_D16)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h
index 0f79e019300..93240d8f1a5 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h
@@ -494,7 +494,6 @@ enum NodeType : unsigned {
   STORE_MSKOR,
   LOAD_CONSTANT,
   TBUFFER_STORE_FORMAT,
-  TBUFFER_STORE_FORMAT_X3,
   TBUFFER_STORE_FORMAT_D16,
   TBUFFER_LOAD_FORMAT,
   TBUFFER_LOAD_FORMAT_D16,
diff --git a/llvm/lib/Target/AMDGPU/BUFInstructions.td b/llvm/lib/Target/AMDGPU/BUFInstructions.td
index d720eb9d46f..030fd459b0f 100644
--- a/llvm/lib/Target/AMDGPU/BUFInstructions.td
+++ b/llvm/lib/Target/AMDGPU/BUFInstructions.td
@@ -1011,11 +1011,11 @@ def BUFFER_WBINVL1 : MUBUF_Invalidate <"buffer_wbinvl1",
 
 defm TBUFFER_LOAD_FORMAT_X     : MTBUF_Pseudo_Loads  <"tbuffer_load_format_x",     VGPR_32>;
 defm TBUFFER_LOAD_FORMAT_XY    : MTBUF_Pseudo_Loads  <"tbuffer_load_format_xy",    VReg_64>;
-defm TBUFFER_LOAD_FORMAT_XYZ   : MTBUF_Pseudo_Loads  <"tbuffer_load_format_xyz",   VReg_128>;
+defm TBUFFER_LOAD_FORMAT_XYZ   : MTBUF_Pseudo_Loads  <"tbuffer_load_format_xyz",   VReg_96>;
 defm TBUFFER_LOAD_FORMAT_XYZW  : MTBUF_Pseudo_Loads  <"tbuffer_load_format_xyzw",  VReg_128>;
 defm TBUFFER_STORE_FORMAT_X    : MTBUF_Pseudo_Stores <"tbuffer_store_format_x",    VGPR_32>;
 defm TBUFFER_STORE_FORMAT_XY   : MTBUF_Pseudo_Stores <"tbuffer_store_format_xy",   VReg_64>;
-defm TBUFFER_STORE_FORMAT_XYZ  : MTBUF_Pseudo_Stores <"tbuffer_store_format_xyz",  VReg_128>;
+defm TBUFFER_STORE_FORMAT_XYZ  : MTBUF_Pseudo_Stores <"tbuffer_store_format_xyz",  VReg_96>;
 defm TBUFFER_STORE_FORMAT_XYZW : MTBUF_Pseudo_Stores <"tbuffer_store_format_xyzw", VReg_128>;
 
 let SubtargetPredicate = HasUnpackedD16VMem, D16Buf = 1 in {
@@ -1104,6 +1104,8 @@ defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, f32, "BUFFER_LOAD_FORMAT_X">
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, i32, "BUFFER_LOAD_FORMAT_X">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, v2f32, "BUFFER_LOAD_FORMAT_XY">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, v2i32, "BUFFER_LOAD_FORMAT_XY">;
+defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, v3f32, "BUFFER_LOAD_FORMAT_XYZ">;
+defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, v3i32, "BUFFER_LOAD_FORMAT_XYZ">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, v4f32, "BUFFER_LOAD_FORMAT_XYZW">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, v4i32, "BUFFER_LOAD_FORMAT_XYZW">;
 
@@ -1127,6 +1129,8 @@ defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, f32, "BUFFER_LOAD_DWORD">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, i32, "BUFFER_LOAD_DWORD">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2f32, "BUFFER_LOAD_DWORDX2">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v2i32, "BUFFER_LOAD_DWORDX2">;
+defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v3f32, "BUFFER_LOAD_DWORDX3">;
+defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v3i32, "BUFFER_LOAD_DWORDX3">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v4f32, "BUFFER_LOAD_DWORDX4">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, v4i32, "BUFFER_LOAD_DWORDX4">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_byte, i32, "BUFFER_LOAD_SBYTE">;
@@ -1172,6 +1176,8 @@ defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, f32, "BUFFER_STORE_FORMAT_
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, i32, "BUFFER_STORE_FORMAT_X">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v2f32, "BUFFER_STORE_FORMAT_XY">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v2i32, "BUFFER_STORE_FORMAT_XY">;
+defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v3f32, "BUFFER_STORE_FORMAT_XYZ">;
+defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v3i32, "BUFFER_STORE_FORMAT_XYZ">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v4f32, "BUFFER_STORE_FORMAT_XYZW">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v4i32, "BUFFER_STORE_FORMAT_XYZW">;
 
@@ -1195,6 +1201,8 @@ defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, f32, "BUFFER_STORE_DWORD">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, i32, "BUFFER_STORE_DWORD">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2f32, "BUFFER_STORE_DWORDX2">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v2i32, "BUFFER_STORE_DWORDX2">;
+defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v3f32, "BUFFER_STORE_DWORDX3">;
+defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v3i32, "BUFFER_STORE_DWORDX3">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v4f32, "BUFFER_STORE_DWORDX4">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, v4i32, "BUFFER_STORE_DWORDX4">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_byte, i32, "BUFFER_STORE_BYTE">;
@@ -1535,9 +1543,11 @@ multiclass MTBUF_LoadIntrinsicPat<SDPatternOperator name, ValueType vt,
 
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, i32,   "TBUFFER_LOAD_FORMAT_X">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v2i32, "TBUFFER_LOAD_FORMAT_XY">;
+defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v3i32, "TBUFFER_LOAD_FORMAT_XYZ">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v4i32, "TBUFFER_LOAD_FORMAT_XYZW">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, f32,   "TBUFFER_LOAD_FORMAT_X">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v2f32, "TBUFFER_LOAD_FORMAT_XY">;
+defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v3f32, "TBUFFER_LOAD_FORMAT_XYZ">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v4f32, "TBUFFER_LOAD_FORMAT_XYZW">;
 
 let SubtargetPredicate = HasUnpackedD16VMem in {
@@ -1591,11 +1601,11 @@ multiclass MTBUF_StoreIntrinsicPat<SDPatternOperator name, ValueType vt,
 
 defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, i32,   "TBUFFER_STORE_FORMAT_X">;
 defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, v2i32, "TBUFFER_STORE_FORMAT_XY">;
-defm : MTBUF_StoreIntrinsicPat<SItbuffer_store_x3, v4i32, "TBUFFER_STORE_FORMAT_XYZ">;
+defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, v3i32, "TBUFFER_STORE_FORMAT_XYZ">;
 defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, v4i32, "TBUFFER_STORE_FORMAT_XYZW">;
 defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, f32,   "TBUFFER_STORE_FORMAT_X">;
 defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, v2f32, "TBUFFER_STORE_FORMAT_XY">;
-defm : MTBUF_StoreIntrinsicPat<SItbuffer_store_x3, v4f32, "TBUFFER_STORE_FORMAT_XYZ">;
+defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, v3f32, "TBUFFER_STORE_FORMAT_XYZ">;
 defm : MTBUF_StoreIntrinsicPat<SItbuffer_store, v4f32, "TBUFFER_STORE_FORMAT_XYZW">;
 
 let SubtargetPredicate = HasUnpackedD16VMem in {
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 7280edd3fcc..4155a013ad8 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -5625,8 +5625,8 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
         LoadVT.getScalarType() == MVT::i16)
       return handleByteShortBufferLoads(DAG, LoadVT, DL, Ops, M);
 
-    return DAG.getMemIntrinsicNode(Opc, DL, Op->getVTList(), Ops, IntVT,
-                                   M->getMemOperand());
+    return getMemIntrinsicNode(Opc, DL, Op->getVTList(), Ops, IntVT,
+                               M->getMemOperand(), DAG);
   }
   case Intrinsic::amdgcn_raw_buffer_load:
   case Intrinsic::amdgcn_raw_buffer_load_format: {
@@ -5659,8 +5659,8 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
         LoadVT.getScalarType() == MVT::i16)
       return handleByteShortBufferLoads(DAG, LoadVT, DL, Ops, M);
 
-    return DAG.getMemIntrinsicNode(Opc, DL, Op->getVTList(), Ops, IntVT,
-                                   M->getMemOperand());
+    return getMemIntrinsicNode(Opc, DL, Op->getVTList(), Ops, IntVT,
+                               M->getMemOperand(), DAG);
   }
   case Intrinsic::amdgcn_struct_buffer_load:
   case Intrinsic::amdgcn_struct_buffer_load_format: {
@@ -5693,8 +5693,8 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
         LoadVT.getScalarType() == MVT::i16)
       return handleByteShortBufferLoads(DAG, LoadVT, DL, Ops, M);
 
-    return DAG.getMemIntrinsicNode(Opc, DL, Op->getVTList(), Ops, IntVT,
-                                   M->getMemOperand());
+    return getMemIntrinsicNode(Opc, DL, Op->getVTList(), Ops, IntVT,
+                               M->getMemOperand(), DAG);
   }
   case Intrinsic::amdgcn_tbuffer_load: {
     MemSDNode *M = cast<MemSDNode>(Op);
@@ -5722,9 +5722,9 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
     if (LoadVT.getScalarType() == MVT::f16)
       return adjustLoadValueType(AMDGPUISD::TBUFFER_LOAD_FORMAT_D16,
                                  M, DAG, Ops);
-    return DAG.getMemIntrinsicNode(AMDGPUISD::TBUFFER_LOAD_FORMAT, DL,
-                                   Op->getVTList(), Ops, LoadVT,
-                                   M->getMemOperand());
+    return getMemIntrinsicNode(AMDGPUISD::TBUFFER_LOAD_FORMAT, DL,
+                               Op->getVTList(), Ops, LoadVT, M->getMemOperand(),
+                               DAG);
   }
   case Intrinsic::amdgcn_raw_tbuffer_load: {
     MemSDNode *M = cast<MemSDNode>(Op);
@@ -5746,9 +5746,9 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
     if (LoadVT.getScalarType() == MVT::f16)
       return adjustLoadValueType(AMDGPUISD::TBUFFER_LOAD_FORMAT_D16,
                                  M, DAG, Ops);
-    return DAG.getMemIntrinsicNode(AMDGPUISD::TBUFFER_LOAD_FORMAT, DL,
-                                   Op->getVTList(), Ops, LoadVT,
-                                   M->getMemOperand());
+    return getMemIntrinsicNode(AMDGPUISD::TBUFFER_LOAD_FORMAT, DL,
+                               Op->getVTList(), Ops, LoadVT, M->getMemOperand(),
+                               DAG);
   }
   case Intrinsic::amdgcn_struct_tbuffer_load: {
     MemSDNode *M = cast<MemSDNode>(Op);
@@ -5770,9 +5770,9 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
     if (LoadVT.getScalarType() == MVT::f16)
       return adjustLoadValueType(AMDGPUISD::TBUFFER_LOAD_FORMAT_D16,
                                  M, DAG, Ops);
-    return DAG.getMemIntrinsicNode(AMDGPUISD::TBUFFER_LOAD_FORMAT, DL,
-                                   Op->getVTList(), Ops, LoadVT,
-                                   M->getMemOperand());
+    return getMemIntrinsicNode(AMDGPUISD::TBUFFER_LOAD_FORMAT, DL,
+                               Op->getVTList(), Ops, LoadVT, M->getMemOperand(),
+                               DAG);
   }
   case Intrinsic::amdgcn_buffer_atomic_swap:
   case Intrinsic::amdgcn_buffer_atomic_add:
@@ -6047,6 +6047,39 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
   }
 }
 
+// Call DAG.getMemIntrinsicNode for a load, but first widen a dwordx3 type to
+// dwordx4 if on SI.
+SDValue SITargetLowering::getMemIntrinsicNode(unsigned Opcode, const SDLoc &DL,
+                                              SDVTList VTList,
+                                              ArrayRef<SDValue> Ops, EVT MemVT,
+                                              MachineMemOperand *MMO,
+                                              SelectionDAG &DAG) const {
+  EVT VT = VTList.VTs[0];
+  EVT WidenedVT = VT;
+  EVT WidenedMemVT = MemVT;
+  if (!Subtarget->hasDwordx3LoadStores() &&
+      (WidenedVT == MVT::v3i32 || WidenedVT == MVT::v3f32)) {
+    WidenedVT = EVT::getVectorVT(*DAG.getContext(),
+                                 WidenedVT.getVectorElementType(), 4);
+    WidenedMemVT = EVT::getVectorVT(*DAG.getContext(),
+                                    WidenedMemVT.getVectorElementType(), 4);
+    MMO = DAG.getMachineFunction().getMachineMemOperand(MMO, 0, 16);
+  }
+
+  assert(VTList.NumVTs == 2);
+  SDVTList WidenedVTList = DAG.getVTList(WidenedVT, VTList.VTs[1]);
+
+  auto NewOp = DAG.getMemIntrinsicNode(Opcode, DL, WidenedVTList, Ops,
+                                       WidenedMemVT, MMO);
+  if (WidenedVT != VT) {
+    auto Extract = DAG.getNode(
+        ISD::EXTRACT_SUBVECTOR, DL, VT, NewOp,
+        DAG.getConstant(0, DL, getVectorIdxTy(DAG.getDataLayout())));
+    NewOp = DAG.getMergeValues({ Extract, SDValue(NewOp.getNode(), 1) }, DL);
+  }
+  return NewOp;
+}
+
 SDValue SITargetLowering::handleD16VData(SDValue VData,
                                          SelectionDAG &DAG) const {
   EVT StoreVT = VData.getValueType();
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.h b/llvm/lib/Target/AMDGPU/SIISelLowering.h
index 01684867cb0..b75f609a507 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.h
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.h
@@ -94,6 +94,12 @@ private:
                               SelectionDAG &DAG, ArrayRef<SDValue> Ops,
                               bool IsIntrinsic = false) const;
 
+  // Call DAG.getMemIntrinsicNode for a load, but first widen a dwordx3 type to
+  // dwordx4 if on SI.
+  SDValue getMemIntrinsicNode(unsigned Opcode, const SDLoc &DL, SDVTList VTList,
+                              ArrayRef<SDValue> Ops, EVT MemVT,
+                              MachineMemOperand *MMO, SelectionDAG &DAG) const;
+
   SDValue handleD16VData(SDValue VData, SelectionDAG &DAG) const;
 
   /// Converts \p Op, which must be of floating point type, to the
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index e10c45cf9ea..13903d70ec5 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -108,9 +108,6 @@ def SDTtbuffer_store : SDTypeProfile<0, 9,
 
 def SItbuffer_store : SDNode<"AMDGPUISD::TBUFFER_STORE_FORMAT", SDTtbuffer_store,
                              [SDNPMayStore, SDNPMemOperand, SDNPHasChain]>;
-def SItbuffer_store_x3 : SDNode<"AMDGPUISD::TBUFFER_STORE_FORMAT_X3",
-                                SDTtbuffer_store,
-                                [SDNPMayStore, SDNPMemOperand, SDNPHasChain]>;
 def SItbuffer_store_d16 : SDNode<"AMDGPUISD::TBUFFER_STORE_FORMAT_D16",
                                 SDTtbuffer_store,
                                 [SDNPMayStore, SDNPMemOperand, SDNPHasChain]>;