https://github.com/MDevereau updated
https://github.com/llvm/llvm-project/pull/72849
>From 3b1d34afb1ae365f48716ae5eb9202a474adf234 Mon Sep 17 00:00:00 2001
From: Matt Devereau <matthew.dever...@arm.com>
Date: Mon, 20 Nov 2023 10:49:27 +0000
Subject: [PATCH] [AArch64][SME2] Add ldr_zt, str_zt builtins and intrinsics
---
clang/include/clang/Basic/arm_sme.td | 8 +++
.../acle_sme2_ldr_str_zt.c | 51 +++++++++++++++++++
.../Target/AArch64/AArch64ISelDAGToDAG.cpp | 8 ++-
.../Target/AArch64/AArch64ISelLowering.cpp | 20 ++++++++
llvm/lib/Target/AArch64/AArch64ISelLowering.h | 2 +
.../Target/AArch64/AArch64RegisterInfo.cpp | 6 +++
.../lib/Target/AArch64/AArch64SMEInstrInfo.td | 4 +-
llvm/lib/Target/AArch64/SMEInstrFormats.td | 23 +++++++--
.../CodeGen/AArch64/sme2-intrinsics-zt0.ll | 27 ++++++++++
9 files changed, 140 insertions(+), 9 deletions(-)
create mode 100644
clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_ldr_str_zt.c
create mode 100644 llvm/test/CodeGen/AArch64/sme2-intrinsics-zt0.ll
diff --git a/clang/include/clang/Basic/arm_sme.td
b/clang/include/clang/Basic/arm_sme.td
index b5655afdf419ecf2..fb3f54ecff95080d 100644
--- a/clang/include/clang/Basic/arm_sme.td
+++ b/clang/include/clang/Basic/arm_sme.td
@@ -298,3 +298,11 @@ multiclass ZAAddSub<string n_suffix> {
defm SVADD : ZAAddSub<"add">;
defm SVSUB : ZAAddSub<"sub">;
+
+//
+// Spill and fill of ZT0
+//
+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/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_ldr_str_zt.c
b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_ldr_str_zt.c
new file mode 100644
index 0000000000000000..3d70ded6b469ba15
--- /dev/null
+++ 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/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
b/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
index 7617dccdeee397f7..abfe14e52509d58a 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
@@ -326,15 +326,19 @@ 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
index d42ae4ff93a4442b..c4b40b7c854f6274 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -2748,6 +2748,22 @@ 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,
@@ -2864,6 +2880,10 @@ 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
index 2a039488f2a9ab36..b690b996d8cddff8 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -612,6 +612,8 @@ 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
index ed64a7b4984c17cd..24ba9dd95004c6f2 100644
--- 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);
if (MF.getFunction().getCallingConv() == CallingConv::GRAAL) {
diff --git a/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td
b/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td
index bb9464a8d2e1cf28..fcfa5f82a3809c2c 100644
--- 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
index 6c9b1f11a4decdee..ef9c323e25bc3585 100644
--- 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_zt : ComplexPattern<i32, 1, "ImmToTile<AArch64::ZT0, 0>",
[]>;
def tileslice8 : ComplexPattern<i32 , 2, "SelectSMETileSlice<15, 1>", []>;
def tileslice16 : ComplexPattern<i32 , 2, "SelectSMETileSlice<7, 1>", []>;
@@ -3137,6 +3138,18 @@ 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 ZTR:$ZTt, GPR64sp:$base), []>, Sched<[]> {
+ // Translated to actual instruction in AArch64ISelLowering.cpp
+ let usesCustomInserter = 1;
+ }
+ def : Pat<(op (imm_to_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
new file mode 100644
index 0000000000000000..30205d86f2fb2034
--- /dev/null
+++ 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)
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
