Skip to content

Commit 8bf31e2

Browse files
sdesmalen-armsdesmalen-arm
committed
[Aarch64][SVE] Add intrinsics for gather loads with 32-bits offsets
This patch adds intrinsics for SVE gather loads for which the offsets are 32-bits wide and are: * unscaled * @llvm.aarch64.sve.ld1.gather.sxtw * @llvm.aarch64.sve.ld1.gather.uxtw * scaled (offsets become indices) * @llvm.arch64.sve.ld1.gather.sxtw.index * @llvm.arch64.sve.ld1.gather.uxtw.index The offsets are either zero (uxtw) or sign (sxtw) extended to 64 bits. These intrinsics map 1-1 to the corresponding SVE instructions (examples for half-words): * unscaled * ld1h { z0.s }, p0/z, [x0, z0.s, sxtw] * ld1h { z0.s }, p0/z, [x0, z0.s, uxtw] * scaled * ld1h { z0.s }, p0/z, [x0, z0.s, sxtw #1] * ld1h { z0.s }, p0/z, [x0, z0.s, uxtw #1] Committed on behalf of Andrzej Warzynski (andwar) Reviewers: sdesmalen, kmclaughlin, eli.friedman, rengolin, rovka, huntergr, dancgr, mgudim, efriedma Reviewed By: sdesmalen Tags: #llvm Differential Revision: https://reviews.llvm.org/D70782
1 parent 8dd17a1 commit 8bf31e2

7 files changed

+573
-46
lines changed

Diff for: ‎llvm/include/llvm/IR/IntrinsicsAArch64.td

+18
Original file line numberDiff line numberDiff line change
@@ -970,6 +970,14 @@ class AdvSIMD_GatherLoad_64bitOffset_Intrinsic
970970
// to reuse currently identical class definitions.
971971
class AdvSIMD_SVE_LOGB_Intrinsic : AdvSIMD_SVE_CNT_Intrinsic;
972972

973+
class AdvSIMD_GatherLoad_32bitOffset_Intrinsic
974+
: Intrinsic<[ llvm_anyvector_ty ],
975+
[
976+
LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
977+
LLVMPointerToElt<0>, llvm_anyvector_ty
978+
],
979+
[ IntrReadMem, IntrArgMemOnly ]>;
980+
973981
// This class of intrinsics are not intended to be useful within LLVM IR but
974982
// are instead here to support some of the more regid parts of the ACLE.
975983
class Builtin_SVCVT<string name, LLVMType OUT, LLVMType IN>
@@ -1211,6 +1219,16 @@ def int_aarch64_sve_ld1_gather : AdvSIMD_GatherLoad_64bitOffset_Intrinsic;
12111219
// scalar + vector, 64 bit scaled offsets
12121220
def int_aarch64_sve_ld1_gather_index : AdvSIMD_GatherLoad_64bitOffset_Intrinsic;
12131221

1222+
// scalar + vector, 32 bit unscaled offsets, sign (sxtw) or zero (zxtw)
1223+
// extended to 64 bits
1224+
def int_aarch64_sve_ld1_gather_sxtw : AdvSIMD_GatherLoad_32bitOffset_Intrinsic;
1225+
def int_aarch64_sve_ld1_gather_uxtw : AdvSIMD_GatherLoad_32bitOffset_Intrinsic;
1226+
1227+
// scalar + vector, 32 bit scaled offsets, sign (sxtw) or zero (zxtw) extended
1228+
// to 64 bits
1229+
def int_aarch64_sve_ld1_gather_sxtw_index : AdvSIMD_GatherLoad_32bitOffset_Intrinsic;
1230+
def int_aarch64_sve_ld1_gather_uxtw_index : AdvSIMD_GatherLoad_32bitOffset_Intrinsic;
1231+
12141232
//
12151233
// SVE2 - Non-widening pairwise arithmetic
12161234
//

Diff for: ‎llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

+12
Original file line numberDiff line numberDiff line change
@@ -1338,6 +1338,10 @@ const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
13381338
case AArch64ISD::INSR: return "AArch64ISD::INSR";
13391339
case AArch64ISD::GLD1: return "AArch64ISD::GLD1";
13401340
case AArch64ISD::GLD1_SCALED: return "AArch64ISD::GLD1_SCALED";
1341+
case AArch64ISD::GLD1_SXTW: return "AArch64ISD::GLD1_SXTW";
1342+
case AArch64ISD::GLD1_UXTW: return "AArch64ISD::GLD1_UXTW";
1343+
case AArch64ISD::GLD1_SXTW_SCALED: return "AArch64ISD::GLD1_SXTW_SCALED";
1344+
case AArch64ISD::GLD1_UXTW_SCALED: return "AArch64ISD::GLD1_UXTW_SCALED";
13411345
}
13421346
return nullptr;
13431347
}
@@ -11931,6 +11935,14 @@ SDValue AArch64TargetLowering::PerformDAGCombine(SDNode *N,
1193111935
return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1);
1193211936
case Intrinsic::aarch64_sve_ld1_gather_index:
1193311937
return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_SCALED);
11938+
case Intrinsic::aarch64_sve_ld1_gather_sxtw:
11939+
return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_SXTW);
11940+
case Intrinsic::aarch64_sve_ld1_gather_uxtw:
11941+
return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_UXTW);
11942+
case Intrinsic::aarch64_sve_ld1_gather_sxtw_index:
11943+
return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_SXTW_SCALED);
11944+
case Intrinsic::aarch64_sve_ld1_gather_uxtw_index:
11945+
return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_UXTW_SCALED);
1193411946
default:
1193511947
break;
1193611948
}

Diff for: ‎llvm/lib/Target/AArch64/AArch64ISelLowering.h

+4
Original file line numberDiff line numberDiff line change
@@ -201,6 +201,10 @@ enum NodeType : unsigned {
201201
// Unsigned gather loads.
202202
GLD1,
203203
GLD1_SCALED,
204+
GLD1_UXTW,
205+
GLD1_SXTW,
206+
GLD1_UXTW_SCALED,
207+
GLD1_SXTW_SCALED,
204208

205209
// NEON Load/Store with post-increment base updates
206210
LD2post = ISD::FIRST_TARGET_MEMORY_OPCODE,

Diff for: ‎llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

+46-42
Original file line numberDiff line numberDiff line change
@@ -15,8 +15,12 @@ def SDT_AArch64_GLD1 : SDTypeProfile<1, 4, [
1515
SDTCVecEltisVT<1,i1>, SDTCisSameNumEltsAs<0,1>
1616
]>;
1717

18-
def AArch64ld1_gather : SDNode<"AArch64ISD::GLD1", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
19-
def AArch64ld1_gather_scaled : SDNode<"AArch64ISD::GLD1_SCALED", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
18+
def AArch64ld1_gather : SDNode<"AArch64ISD::GLD1", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
19+
def AArch64ld1_gather_scaled : SDNode<"AArch64ISD::GLD1_SCALED", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
20+
def AArch64ld1_gather_uxtw : SDNode<"AArch64ISD::GLD1_UXTW", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
21+
def AArch64ld1_gather_sxtw : SDNode<"AArch64ISD::GLD1_SXTW", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
22+
def AArch64ld1_gather_uxtw_scaled : SDNode<"AArch64ISD::GLD1_UXTW_SCALED", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
23+
def AArch64ld1_gather_sxtw_scaled : SDNode<"AArch64ISD::GLD1_SXTW_SCALED", SDT_AArch64_GLD1, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
2024

2125
let Predicates = [HasSVE] in {
2226

@@ -410,25 +414,25 @@ let Predicates = [HasSVE] in {
410414

411415
// Gathers using unscaled 32-bit offsets, e.g.
412416
// ld1h z0.s, p0/z, [x0, z0.s, uxtw]
413-
defm GLD1SB_S : sve_mem_32b_gld_vs_32_unscaled<0b0000, "ld1sb", ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only>;
414-
defm GLDFF1SB_S : sve_mem_32b_gld_vs_32_unscaled<0b0001, "ldff1sb", ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only>;
415-
defm GLD1B_S : sve_mem_32b_gld_vs_32_unscaled<0b0010, "ld1b", ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only>;
416-
defm GLDFF1B_S : sve_mem_32b_gld_vs_32_unscaled<0b0011, "ldff1b", ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only>;
417-
defm GLD1SH_S : sve_mem_32b_gld_vs_32_unscaled<0b0100, "ld1sh", ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
418-
defm GLDFF1SH_S : sve_mem_32b_gld_vs_32_unscaled<0b0101, "ldff1sh", ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
419-
defm GLD1H_S : sve_mem_32b_gld_vs_32_unscaled<0b0110, "ld1h", ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
420-
defm GLDFF1H_S : sve_mem_32b_gld_vs_32_unscaled<0b0111, "ldff1h", ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
421-
defm GLD1W : sve_mem_32b_gld_vs_32_unscaled<0b1010, "ld1w", ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
422-
defm GLDFF1W : sve_mem_32b_gld_vs_32_unscaled<0b1011, "ldff1w", ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
417+
defm GLD1SB_S : sve_mem_32b_gld_vs_32_unscaled<0b0000, "ld1sb", null_frag, null_frag, ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only, nxv4i8>;
418+
defm GLDFF1SB_S : sve_mem_32b_gld_vs_32_unscaled<0b0001, "ldff1sb", null_frag, null_frag, ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only, nxv4i8>;
419+
defm GLD1B_S : sve_mem_32b_gld_vs_32_unscaled<0b0010, "ld1b", AArch64ld1_gather_sxtw, AArch64ld1_gather_uxtw, ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only, nxv4i8>;
420+
defm GLDFF1B_S : sve_mem_32b_gld_vs_32_unscaled<0b0011, "ldff1b", null_frag, null_frag, ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only, nxv4i8>;
421+
defm GLD1SH_S : sve_mem_32b_gld_vs_32_unscaled<0b0100, "ld1sh", null_frag, null_frag, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i16>;
422+
defm GLDFF1SH_S : sve_mem_32b_gld_vs_32_unscaled<0b0101, "ldff1sh", null_frag, null_frag, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i16>;
423+
defm GLD1H_S : sve_mem_32b_gld_vs_32_unscaled<0b0110, "ld1h", AArch64ld1_gather_sxtw, AArch64ld1_gather_uxtw, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i16>;
424+
defm GLDFF1H_S : sve_mem_32b_gld_vs_32_unscaled<0b0111, "ldff1h", null_frag, null_frag, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i16>;
425+
defm GLD1W : sve_mem_32b_gld_vs_32_unscaled<0b1010, "ld1w", AArch64ld1_gather_sxtw, AArch64ld1_gather_uxtw, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i32>;
426+
defm GLDFF1W : sve_mem_32b_gld_vs_32_unscaled<0b1011, "ldff1w", null_frag, null_frag, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i32>;
423427

424428
// Gathers using scaled 32-bit offsets, e.g.
425429
// ld1h z0.s, p0/z, [x0, z0.s, uxtw #1]
426-
defm GLD1SH_S : sve_mem_32b_gld_sv_32_scaled<0b0100, "ld1sh", ZPR32ExtSXTW16, ZPR32ExtUXTW16>;
427-
defm GLDFF1SH_S : sve_mem_32b_gld_sv_32_scaled<0b0101, "ldff1sh", ZPR32ExtSXTW16, ZPR32ExtUXTW16>;
428-
defm GLD1H_S : sve_mem_32b_gld_sv_32_scaled<0b0110, "ld1h", ZPR32ExtSXTW16, ZPR32ExtUXTW16>;
429-
defm GLDFF1H_S : sve_mem_32b_gld_sv_32_scaled<0b0111, "ldff1h", ZPR32ExtSXTW16, ZPR32ExtUXTW16>;
430-
defm GLD1W : sve_mem_32b_gld_sv_32_scaled<0b1010, "ld1w", ZPR32ExtSXTW32, ZPR32ExtUXTW32>;
431-
defm GLDFF1W : sve_mem_32b_gld_sv_32_scaled<0b1011, "ldff1w", ZPR32ExtSXTW32, ZPR32ExtUXTW32>;
430+
defm GLD1SH_S : sve_mem_32b_gld_sv_32_scaled<0b0100, "ld1sh", null_frag, null_frag, ZPR32ExtSXTW16, ZPR32ExtUXTW16, nxv4i16>;
431+
defm GLDFF1SH_S : sve_mem_32b_gld_sv_32_scaled<0b0101, "ldff1sh", null_frag, null_frag, ZPR32ExtSXTW16, ZPR32ExtUXTW16, nxv4i16>;
432+
defm GLD1H_S : sve_mem_32b_gld_sv_32_scaled<0b0110, "ld1h", AArch64ld1_gather_sxtw_scaled, AArch64ld1_gather_uxtw_scaled, ZPR32ExtSXTW16, ZPR32ExtUXTW16, nxv4i16>;
433+
defm GLDFF1H_S : sve_mem_32b_gld_sv_32_scaled<0b0111, "ldff1h", null_frag, null_frag, ZPR32ExtSXTW16, ZPR32ExtUXTW16, nxv4i16>;
434+
defm GLD1W : sve_mem_32b_gld_sv_32_scaled<0b1010, "ld1w", AArch64ld1_gather_sxtw_scaled, AArch64ld1_gather_uxtw_scaled, ZPR32ExtSXTW32, ZPR32ExtUXTW32, nxv4i32>;
435+
defm GLDFF1W : sve_mem_32b_gld_sv_32_scaled<0b1011, "ldff1w", null_frag, null_frag, ZPR32ExtSXTW32, ZPR32ExtUXTW32, nxv4i32>;
432436

433437
// Gathers using scaled 32-bit pointers with offset, e.g.
434438
// ld1h z0.s, p0/z, [z0.s, #16]
@@ -492,33 +496,33 @@ let Predicates = [HasSVE] in {
492496

493497
// Gathers using unscaled 32-bit offsets unpacked in 64-bits elements, e.g.
494498
// ld1h z0.d, p0/z, [x0, z0.d, uxtw]
495-
defm GLD1SB_D : sve_mem_64b_gld_vs_32_unscaled<0b0000, "ld1sb", ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only>;
496-
defm GLDFF1SB_D : sve_mem_64b_gld_vs_32_unscaled<0b0001, "ldff1sb", ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only>;
497-
defm GLD1B_D : sve_mem_64b_gld_vs_32_unscaled<0b0010, "ld1b", ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only>;
498-
defm GLDFF1B_D : sve_mem_64b_gld_vs_32_unscaled<0b0011, "ldff1b", ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only>;
499-
defm GLD1SH_D : sve_mem_64b_gld_vs_32_unscaled<0b0100, "ld1sh", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
500-
defm GLDFF1SH_D : sve_mem_64b_gld_vs_32_unscaled<0b0101, "ldff1sh", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
501-
defm GLD1H_D : sve_mem_64b_gld_vs_32_unscaled<0b0110, "ld1h", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
502-
defm GLDFF1H_D : sve_mem_64b_gld_vs_32_unscaled<0b0111, "ldff1h", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
503-
defm GLD1SW_D : sve_mem_64b_gld_vs_32_unscaled<0b1000, "ld1sw", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
504-
defm GLDFF1SW_D : sve_mem_64b_gld_vs_32_unscaled<0b1001, "ldff1sw", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
505-
defm GLD1W_D : sve_mem_64b_gld_vs_32_unscaled<0b1010, "ld1w", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
506-
defm GLDFF1W_D : sve_mem_64b_gld_vs_32_unscaled<0b1011, "ldff1w", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
507-
defm GLD1D : sve_mem_64b_gld_vs_32_unscaled<0b1110, "ld1d", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
508-
defm GLDFF1D : sve_mem_64b_gld_vs_32_unscaled<0b1111, "ldff1d", ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
499+
defm GLD1SB_D : sve_mem_64b_gld_vs_32_unscaled<0b0000, "ld1sb", null_frag, null_frag, ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only, nxv2i8>;
500+
defm GLDFF1SB_D : sve_mem_64b_gld_vs_32_unscaled<0b0001, "ldff1sb", null_frag, null_frag, ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only, nxv2i8>;
501+
defm GLD1B_D : sve_mem_64b_gld_vs_32_unscaled<0b0010, "ld1b", AArch64ld1_gather_sxtw, AArch64ld1_gather_uxtw, ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only, nxv2i8>;
502+
defm GLDFF1B_D : sve_mem_64b_gld_vs_32_unscaled<0b0011, "ldff1b", null_frag, null_frag, ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only, nxv2i8>;
503+
defm GLD1SH_D : sve_mem_64b_gld_vs_32_unscaled<0b0100, "ld1sh", null_frag, null_frag, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i16>;
504+
defm GLDFF1SH_D : sve_mem_64b_gld_vs_32_unscaled<0b0101, "ldff1sh", null_frag, null_frag, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i16>;
505+
defm GLD1H_D : sve_mem_64b_gld_vs_32_unscaled<0b0110, "ld1h", AArch64ld1_gather_sxtw, AArch64ld1_gather_uxtw, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i16>;
506+
defm GLDFF1H_D : sve_mem_64b_gld_vs_32_unscaled<0b0111, "ldff1h", null_frag, null_frag, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i16>;
507+
defm GLD1SW_D : sve_mem_64b_gld_vs_32_unscaled<0b1000, "ld1sw", null_frag, null_frag, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i32>;
508+
defm GLDFF1SW_D : sve_mem_64b_gld_vs_32_unscaled<0b1001, "ldff1sw", null_frag, null_frag, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i32>;
509+
defm GLD1W_D : sve_mem_64b_gld_vs_32_unscaled<0b1010, "ld1w", AArch64ld1_gather_sxtw, AArch64ld1_gather_uxtw, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i32>;
510+
defm GLDFF1W_D : sve_mem_64b_gld_vs_32_unscaled<0b1011, "ldff1w", null_frag, null_frag, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i32>;
511+
defm GLD1D : sve_mem_64b_gld_vs_32_unscaled<0b1110, "ld1d", AArch64ld1_gather_sxtw, AArch64ld1_gather_uxtw, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i64>;
512+
defm GLDFF1D : sve_mem_64b_gld_vs_32_unscaled<0b1111, "ldff1d", null_frag, null_frag, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i64>;
509513

510514
// Gathers using scaled 32-bit offsets unpacked in 64-bits elements, e.g.
511515
// ld1h z0.d, p0/z, [x0, z0.d, uxtw #1]
512-
defm GLD1SH_D : sve_mem_64b_gld_sv_32_scaled<0b0100, "ld1sh", ZPR64ExtSXTW16, ZPR64ExtUXTW16>;
513-
defm GLDFF1SH_D : sve_mem_64b_gld_sv_32_scaled<0b0101, "ldff1sh",ZPR64ExtSXTW16, ZPR64ExtUXTW16>;
514-
defm GLD1H_D : sve_mem_64b_gld_sv_32_scaled<0b0110, "ld1h", ZPR64ExtSXTW16, ZPR64ExtUXTW16>;
515-
defm GLDFF1H_D : sve_mem_64b_gld_sv_32_scaled<0b0111, "ldff1h", ZPR64ExtSXTW16, ZPR64ExtUXTW16>;
516-
defm GLD1SW_D : sve_mem_64b_gld_sv_32_scaled<0b1000, "ld1sw", ZPR64ExtSXTW32, ZPR64ExtUXTW32>;
517-
defm GLDFF1SW_D : sve_mem_64b_gld_sv_32_scaled<0b1001, "ldff1sw",ZPR64ExtSXTW32, ZPR64ExtUXTW32>;
518-
defm GLD1W_D : sve_mem_64b_gld_sv_32_scaled<0b1010, "ld1w", ZPR64ExtSXTW32, ZPR64ExtUXTW32>;
519-
defm GLDFF1W_D : sve_mem_64b_gld_sv_32_scaled<0b1011, "ldff1w", ZPR64ExtSXTW32, ZPR64ExtUXTW32>;
520-
defm GLD1D : sve_mem_64b_gld_sv_32_scaled<0b1110, "ld1d", ZPR64ExtSXTW64, ZPR64ExtUXTW64>;
521-
defm GLDFF1D : sve_mem_64b_gld_sv_32_scaled<0b1111, "ldff1d", ZPR64ExtSXTW64, ZPR64ExtUXTW64>;
516+
defm GLD1SH_D : sve_mem_64b_gld_sv_32_scaled<0b0100, "ld1sh", null_frag, null_frag, ZPR64ExtSXTW16, ZPR64ExtUXTW16, nxv2i16>;
517+
defm GLDFF1SH_D : sve_mem_64b_gld_sv_32_scaled<0b0101, "ldff1sh", null_frag, null_frag, ZPR64ExtSXTW16, ZPR64ExtUXTW16, nxv2i16>;
518+
defm GLD1H_D : sve_mem_64b_gld_sv_32_scaled<0b0110, "ld1h", AArch64ld1_gather_sxtw_scaled, AArch64ld1_gather_uxtw_scaled, ZPR64ExtSXTW16, ZPR64ExtUXTW16, nxv2i16>;
519+
defm GLDFF1H_D : sve_mem_64b_gld_sv_32_scaled<0b0111, "ldff1h", null_frag, null_frag, ZPR64ExtSXTW16, ZPR64ExtUXTW16, nxv2i16>;
520+
defm GLD1SW_D : sve_mem_64b_gld_sv_32_scaled<0b1000, "ld1sw", null_frag, null_frag, ZPR64ExtSXTW32, ZPR64ExtUXTW32, nxv2i32>;
521+
defm GLDFF1SW_D : sve_mem_64b_gld_sv_32_scaled<0b1001, "ldff1sw", null_frag, null_frag, ZPR64ExtSXTW32, ZPR64ExtUXTW32, nxv2i32>;
522+
defm GLD1W_D : sve_mem_64b_gld_sv_32_scaled<0b1010, "ld1w", AArch64ld1_gather_sxtw_scaled, AArch64ld1_gather_uxtw_scaled, ZPR64ExtSXTW32, ZPR64ExtUXTW32, nxv2i32>;
523+
defm GLDFF1W_D : sve_mem_64b_gld_sv_32_scaled<0b1011, "ldff1w", null_frag, null_frag, ZPR64ExtSXTW32, ZPR64ExtUXTW32, nxv2i32>;
524+
defm GLD1D : sve_mem_64b_gld_sv_32_scaled<0b1110, "ld1d", AArch64ld1_gather_sxtw_scaled, AArch64ld1_gather_uxtw_scaled, ZPR64ExtSXTW64, ZPR64ExtUXTW64, nxv2i64>;
525+
defm GLDFF1D : sve_mem_64b_gld_sv_32_scaled<0b1111, "ldff1d", null_frag, null_frag, ZPR64ExtSXTW64, ZPR64ExtUXTW64, nxv2i64>;
522526

523527
// Non-temporal contiguous loads (register + immediate)
524528
defm LDNT1B_ZRI : sve_mem_cldnt_si<0b00, "ldnt1b", Z_b, ZPR8>;

0 commit comments

Comments
 (0)