[AArch64][SME2] Add ldr_zt, str_zt builtins and intrinsics

MDevereau · MDevereau · commit 9846bc9efd79 · 2023-11-09T10:56:45.000Z
Adds the builtins:

void svldr_zt(uint64_t zt, const void *rn)
void svstr_zt(uint64_t zt, void *rn)

And the intrinsics:
call void @llvm.aarch64.sme.ldr.zt(i32, ptr)
tail call void @llvm.aarch64.sme.str.zt(i32, ptr)
diff --git a/clang/include/clang/Basic/arm_sme.td b/clang/include/clang/Basic/arm_sme.td
@@ -298,3 +298,8 @@ multiclass ZAAddSub<string n_suffix> {
 
 defm SVADD : ZAAddSub<"add">;
 defm SVSUB : ZAAddSub<"sub">;
+
+let TargetGuard = "sme2" in {
+  def SVLDR_ZT : Inst<"svldr_zt", "viQ", "", MergeNone, "aarch64_sme_ldr_zt", [IsOverloadNone, IsStreamingCompatible, IsSharedZA, IsPreservesZA], [ImmCheck<0, ImmCheck0_0>]>;
+  def SVSTR_ZT : Inst<"svstr_zt", "vi%", "", MergeNone, "aarch64_sme_str_zt", [IsOverloadNone, IsStreamingCompatible, IsSharedZA, IsPreservesZA], [ImmCheck<0, ImmCheck0_0>]>;
+}
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
@@ -1813,6 +1813,15 @@ def SVWHILERW_H_BF16 : SInst<"svwhilerw[_{1}]", "Pcc", "b", MergeNone, "aarch64_
 def SVWHILEWR_H_BF16 : SInst<"svwhilewr[_{1}]", "Pcc", "b", MergeNone, "aarch64_sve_whilewr_h", [IsOverloadWhileRW]>;
 }
 
+// //
+// // Spill and fill of ZT0
+// //
+
+// let TargetGuard = "sme2" in {
+//   def SVLDR_ZT : Inst<"svldr_zt", "viQ", "", "aarch64_sme_ldr_zt", [IsOverloadNone, IsStreamingCompatible, IsSharedZA, IsPreservesZA], [ImmCheck<0, ImmCheck0_0>]>;
+//   def SVSTR_ZT : Inst<"svstr_zt", "vi%", "", "aarch64_sme_str_zt", [IsOverloadNone, IsStreamingCompatible, IsSharedZA, IsPreservesZA], [ImmCheck<0, ImmCheck0_0>]>;
+// }
+
 ////////////////////////////////////////////////////////////////////////////////
 // SVE2 - Extended table lookup/permute
 let TargetGuard = "sve2" in {
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_ldr_str_zt.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_ldr_str_zt.c
@@ -0,0 +1,51 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+
+// REQUIRES: aarch64-registered-target
+
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+
+#include <arm_sme_draft_spec_subject_to_change.h>
+
+#ifdef SVE_OVERLOADED_FORMS
+// A simple used,unused... macro, long enough to represent any SVE builtin.
+#define SVE_ACLE_FUNC(A1,A2_UNUSED) A1
+#else
+#define SVE_ACLE_FUNC(A1,A2) A1##A2
+#endif
+
+// LDR ZT0
+
+// CHECK-LABEL: @test_svldr_zt(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.ldr.zt(i32 0, ptr [[BASE:%.*]])
+// CHECK-NEXT:    ret void
+//
+// CPP-CHECK-LABEL: @_Z13test_svldr_ztPKv(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.ldr.zt(i32 0, ptr [[BASE:%.*]])
+// CPP-CHECK-NEXT:    ret void
+//
+void test_svldr_zt(const void *base) __arm_streaming_compatible __arm_shared_za __arm_preserves_za {
+  svldr_zt(0, base);
+} ;
+
+
+// STR ZT0
+
+// CHECK-LABEL: @test_svstr_zt(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.str.zt(i32 0, ptr [[BASE:%.*]])
+// CHECK-NEXT:    ret void
+//
+// CPP-CHECK-LABEL: @_Z13test_svstr_ztPv(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.str.zt(i32 0, ptr [[BASE:%.*]])
+// CPP-CHECK-NEXT:    ret void
+//
+void test_svstr_zt(void *base) __arm_streaming_compatible __arm_shared_za __arm_preserves_za {
+  svstr_zt(0, base);
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsAArch64.td b/llvm/include/llvm/IR/IntrinsicsAArch64.td
@@ -2679,10 +2679,10 @@ let TargetPrefix = "aarch64" in {
   def int_aarch64_sme_st1q_vert  : SME_Load_Store_Intrinsic<llvm_nxv1i1_ty>;
 
   // Spill + fill
-  def int_aarch64_sme_ldr : DefaultAttrsIntrinsic<
-    [], [llvm_i32_ty, llvm_ptr_ty]>;
-  def int_aarch64_sme_str : DefaultAttrsIntrinsic<
-    [], [llvm_i32_ty, llvm_ptr_ty]>;
+  class SME_LDR_STR_Intrinsic
+    : DefaultAttrsIntrinsic<[], [llvm_i32_ty, llvm_ptr_ty]>;
+  def int_aarch64_sme_ldr : SME_LDR_STR_Intrinsic;
+  def int_aarch64_sme_str : SME_LDR_STR_Intrinsic;
 
   class SME_TileToVector_Intrinsic
       : DefaultAttrsIntrinsic<[llvm_anyvector_ty],
@@ -3454,4 +3454,7 @@ let TargetPrefix = "aarch64" in {
   def int_aarch64_sve_sel_x2  : SVE2_VG2_Sel_Intrinsic;
   def int_aarch64_sve_sel_x4  : SVE2_VG4_Sel_Intrinsic;
 
+  def int_aarch64_sme_ldr_zt : SME_LDR_STR_Intrinsic;
+  def int_aarch64_sme_str_zt : SME_LDR_STR_Intrinsic;
+
 }
diff --git a/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp b/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
@@ -326,15 +326,18 @@ class AArch64DAGToDAGISel : public SelectionDAGISel {
     return false;
   }
 
-  template <unsigned BaseReg> bool ImmToTile(SDValue N, SDValue &Imm) {
+  template <unsigned BaseReg, unsigned Max> bool ImmToTile(SDValue N, SDValue &Imm) {
     if (auto *CI = dyn_cast<ConstantSDNode>(N)) {
       uint64_t C = CI->getZExtValue();
+
+      if (C > Max)
+        return false;
+
       Imm = CurDAG->getRegister(BaseReg + C, MVT::Other);
       return true;
     }
     return false;
   }
-
   /// Form sequences of consecutive 64/128-bit registers for use in NEON
   /// instructions making use of a vector-list (e.g. ldN, tbl). Vecs must have
   /// between 1 and 4 elements. If it contains a single element that is returned
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -2746,6 +2746,23 @@ AArch64TargetLowering::EmitFill(MachineInstr &MI, MachineBasicBlock *BB) const {
   return BB;
 }
 
+MachineBasicBlock *AArch64TargetLowering::EmitZTSpillFill(MachineInstr &MI,
+                                                          MachineBasicBlock *BB,
+                                                          bool IsSpill) const {
+  const TargetInstrInfo *TII = Subtarget->getInstrInfo();
+  MachineInstrBuilder MIB;
+  if (IsSpill) {
+    MIB = BuildMI(*BB, MI, MI.getDebugLoc(), TII->get(AArch64::STR_TX));
+    MIB.addReg(MI.getOperand(0).getReg());
+  }
+  else
+    MIB = BuildMI(*BB, MI, MI.getDebugLoc(), TII->get(AArch64::LDR_TX),
+                  MI.getOperand(0).getReg());
+  MIB.add(MI.getOperand(1)); // Base
+  MI.eraseFromParent(); // The pseudo is gone now.
+  return BB;
+}
+
 MachineBasicBlock *
 AArch64TargetLowering::EmitZAInstr(unsigned Opc, unsigned BaseReg,
                                    MachineInstr &MI,
@@ -2862,6 +2879,10 @@ MachineBasicBlock *AArch64TargetLowering::EmitInstrWithCustomInserter(
     return EmitTileLoad(AArch64::LD1_MXIPXX_V_Q, AArch64::ZAQ0, MI, BB);
   case AArch64::LDR_ZA_PSEUDO:
     return EmitFill(MI, BB);
+  case AArch64::LDR_TX_PSEUDO:
+    return EmitZTSpillFill(MI, BB, /*IsSpill=*/false);
+  case AArch64::STR_TX_PSEUDO:
+    return EmitZTSpillFill(MI, BB, /*IsSpill=*/true);
   case AArch64::ZERO_M_PSEUDO:
     return EmitZero(MI, BB);
   }
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -608,6 +608,8 @@ class AArch64TargetLowering : public TargetLowering {
   MachineBasicBlock *EmitZAInstr(unsigned Opc, unsigned BaseReg,
                                  MachineInstr &MI, MachineBasicBlock *BB,
                                  bool HasTile) const;
+  MachineBasicBlock *EmitZTSpillFill(MachineInstr &MI, MachineBasicBlock *BB,
+                                     bool IsSpill) const;
   MachineBasicBlock *EmitZero(MachineInstr &MI, MachineBasicBlock *BB) const;
 
   MachineBasicBlock *
diff --git a/llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp b/llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp
@@ -440,6 +440,12 @@ AArch64RegisterInfo::getStrictlyReservedRegs(const MachineFunction &MF) const {
       Reserved.set(SubReg);
   }
 
+  if (MF.getSubtarget<AArch64Subtarget>().hasSME2()) {
+    for (MCSubRegIterator SubReg(AArch64::ZT0, this, /*self=*/true);
+      SubReg.isValid(); ++SubReg)
+      Reserved.set(*SubReg);
+  }
+
   markSuperRegs(Reserved, AArch64::FPCR);
 
   assert(checkAllSuperRegsMarked(Reserved));
diff --git a/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td
@@ -541,8 +541,8 @@ defm UMOPS_MPPZZ_HtoS : sme2_int_mopx_tile<"umops", 0b101, int_aarch64_sme_umops
 
 def ZERO_T : sme2_zero_zt<"zero", 0b0001>;
 
-def LDR_TX : sme2_spill_fill_vector<"ldr", 0b01111100>;
-def STR_TX : sme2_spill_fill_vector<"str", 0b11111100>;
+defm LDR_TX : sme2_spill_fill_vector<"ldr", 0b01111100, int_aarch64_sme_ldr_zt>;
+defm STR_TX : sme2_spill_fill_vector<"str", 0b11111100, int_aarch64_sme_str_zt>;
 
 def MOVT_XTI : sme2_movt_zt_to_scalar<"movt", 0b0011111>;
 def MOVT_TIX : sme2_movt_scalar_to_zt<"movt", 0b0011111>;
diff --git a/llvm/lib/Target/AArch64/SMEInstrFormats.td b/llvm/lib/Target/AArch64/SMEInstrFormats.td
@@ -10,11 +10,12 @@
 //
 //===----------------------------------------------------------------------===//
 
-def imm_to_tile8   : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAB0>", []>;
-def imm_to_tile16  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAH0>", []>;
-def imm_to_tile32  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAS0>", []>;
-def imm_to_tile64  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAD0>", []>;
-def imm_to_tile128 : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAQ0>", []>;
+def imm_to_tile8   : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAB0, 0>",  []>;
+def imm_to_tile16  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAH0, 1>",  []>;
+def imm_to_tile32  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAS0, 3>",  []>;
+def imm_to_tile64  : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAD0, 7>",  []>;
+def imm_to_tile128 : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZAQ0, 15>", []>;
+def imm_to_tile_zt : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZT0,  0>",  []>;
 
 def tileslice8   : ComplexPattern<i32 , 2, "SelectSMETileSlice<15, 1>", []>;
 def tileslice16  : ComplexPattern<i32 , 2, "SelectSMETileSlice<7,  1>", []>;
@@ -3132,6 +3133,18 @@ class sme2_spill_fill_vector<string mnemonic, bits<8> opc>
   let mayStore    = opc{7};
 }
 
+
+multiclass sme2_spill_fill_vector<string mnemonic, bits<8> opc, SDPatternOperator op> {
+  def NAME : sme2_spill_fill_vector<mnemonic, opc>;
+  def NAME # _PSEUDO
+      : Pseudo<(outs), (ins MatrixOp:$ZTt, GPR64sp:$base), []>, Sched<[]> {
+    // Translated to actual instruction in AArch64ISelLowering.cpp
+    let usesCustomInserter = 1;
+  }
+  def : Pat<(op (imm_to_tile_zt untyped:$tile), GPR64sp:$base),
+            (!cast<Instruction>(NAME # _PSEUDO) $tile, $base)>;
+}
+
 //===----------------------------------------------------------------------===///
 // SME2 move to/from lookup table
 class sme2_movt_zt_to_scalar<string mnemonic, bits<7> opc>
diff --git a/llvm/test/CodeGen/AArch64/sme2-intrinsics-zt0.ll b/llvm/test/CodeGen/AArch64/sme2-intrinsics-zt0.ll
@@ -0,0 +1,27 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2 -verify-machineinstrs < %s | FileCheck %s
+
+; LDR
+
+define void @ldr_zt0(ptr %ptr) {
+; CHECK-LABEL: ldr_zt0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr zt0, [x0]
+; CHECK-NEXT:    ret
+  call void @llvm.aarch64.sme.ldr.zt(i32 0, ptr %ptr)
+  ret void;
+}
+
+; STR
+
+define void @str_zt0(ptr %ptr) {
+; CHECK-LABEL: str_zt0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    str zt0, [x0]
+; CHECK-NEXT:    ret
+  call void @llvm.aarch64.sme.str.zt(i32 0, ptr %ptr)
+  ret void;
+}
+
+declare void @llvm.aarch64.sme.ldr.zt(i32, ptr)
+declare void @llvm.aarch64.sme.str.zt(i32, ptr)
