Skip to content

Commit 42b28c6

Browse files
SamTebbs33SamTebbs33
authored
[clang] Add missing streaming attributes to SVE builtins (#79134)
This patch adds `IsStreamingCompatible` or `IsStreamingOrSVE2p1` to the SVE builtins that missed them.
1 parent 51f9e98 commit 42b28c6

File tree

2 files changed

+83
-77
lines changed

2 files changed

+83
-77
lines changed

‎clang/include/clang/Basic/arm_sve.td

Lines changed: 43 additions & 43 deletions
Original file line numberDiff line numberDiff line change
@@ -543,13 +543,13 @@ def SVADRD : SInst<"svadrd[_{0}base]_[{2}]index", "uud", "ilUiUl", MergeNone, "
543543
////////////////////////////////////////////////////////////////////////////////
544544
// Scalar to vector
545545

546-
def SVDUPQ_8 : SInst<"svdupq[_n]_{d}", "dssssssssssssssss", "cUc", MergeNone>;
547-
def SVDUPQ_16 : SInst<"svdupq[_n]_{d}", "dssssssss", "sUsh", MergeNone>;
546+
def SVDUPQ_8 : SInst<"svdupq[_n]_{d}", "dssssssssssssssss", "cUc", MergeNone, "", [IsStreamingCompatible]>;
547+
def SVDUPQ_16 : SInst<"svdupq[_n]_{d}", "dssssssss", "sUsh", MergeNone, "", [IsStreamingCompatible]>;
548548
let TargetGuard = "sve,bf16" in {
549-
def SVDUPQ_BF16 : SInst<"svdupq[_n]_{d}", "dssssssss", "b", MergeNone>;
549+
def SVDUPQ_BF16 : SInst<"svdupq[_n]_{d}", "dssssssss", "b", MergeNone, "", [IsStreamingCompatible]>;
550550
}
551-
def SVDUPQ_32 : SInst<"svdupq[_n]_{d}", "dssss", "iUif", MergeNone>;
552-
def SVDUPQ_64 : SInst<"svdupq[_n]_{d}", "dss", "lUld", MergeNone>;
551+
def SVDUPQ_32 : SInst<"svdupq[_n]_{d}", "dssss", "iUif", MergeNone, "", [IsStreamingCompatible]>;
552+
def SVDUPQ_64 : SInst<"svdupq[_n]_{d}", "dss", "lUld", MergeNone, "", [IsStreamingCompatible]>;
553553

554554
multiclass svdup_base<string n, string p, MergeType mt, string i> {
555555
def NAME : SInst<n, p, "csilUcUsUiUlhfd", mt, i, [IsStreamingCompatible]>;
@@ -638,8 +638,8 @@ def SVQADD_N_U : SInst<"svqadd[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64
638638
def SVQSUB_N_S : SInst<"svqsub[_n_{d}]", "dda", "csil", MergeNone, "aarch64_sve_sqsub_x", [IsStreamingCompatible]>;
639639
def SVQSUB_N_U : SInst<"svqsub[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_uqsub_x", [IsStreamingCompatible]>;
640640

641-
def SVDOT_LANE_S : SInst<"svdot_lane[_{d}]", "ddqqi", "il", MergeNone, "aarch64_sve_sdot_lane", [], [ImmCheck<3, ImmCheckLaneIndexDot, 2>]>;
642-
def SVDOT_LANE_U : SInst<"svdot_lane[_{d}]", "ddqqi", "UiUl", MergeNone, "aarch64_sve_udot_lane", [], [ImmCheck<3, ImmCheckLaneIndexDot, 2>]>;
641+
def SVDOT_LANE_S : SInst<"svdot_lane[_{d}]", "ddqqi", "il", MergeNone, "aarch64_sve_sdot_lane", [IsStreamingCompatible], [ImmCheck<3, ImmCheckLaneIndexDot, 2>]>;
642+
def SVDOT_LANE_U : SInst<"svdot_lane[_{d}]", "ddqqi", "UiUl", MergeNone, "aarch64_sve_udot_lane", [IsStreamingCompatible], [ImmCheck<3, ImmCheckLaneIndexDot, 2>]>;
643643

644644
////////////////////////////////////////////////////////////////////////////////
645645
// Logical operations
@@ -835,14 +835,14 @@ def SVSCALE_N_M : SInst<"svscale[_n_{d}]", "dPdK", "hfd", MergeOp1, "aarch64_sv
835835
def SVSCALE_N_X : SInst<"svscale[_n_{d}]", "dPdK", "hfd", MergeAny, "aarch64_sve_fscale", [IsStreamingCompatible]>;
836836
def SVSCALE_N_Z : SInst<"svscale[_n_{d}]", "dPdK", "hfd", MergeZero, "aarch64_sve_fscale", [IsStreamingCompatible]>;
837837

838-
defm SVMAD_F : SInstZPZZZ<"svmad", "hfd", "aarch64_sve_fmad", "aarch64_sve_fmla_u", [ReverseMergeAnyAccOp]>;
839-
defm SVMLA_F : SInstZPZZZ<"svmla", "hfd", "aarch64_sve_fmla", "aarch64_sve_fmla_u">;
840-
defm SVMLS_F : SInstZPZZZ<"svmls", "hfd", "aarch64_sve_fmls", "aarch64_sve_fmls_u">;
841-
defm SVMSB_F : SInstZPZZZ<"svmsb", "hfd", "aarch64_sve_fmsb", "aarch64_sve_fmls_u", [ReverseMergeAnyAccOp]>;
842-
defm SVNMAD_F : SInstZPZZZ<"svnmad", "hfd", "aarch64_sve_fnmad", "aarch64_sve_fnmla_u", [ReverseMergeAnyAccOp]>;
843-
defm SVNMLA_F : SInstZPZZZ<"svnmla", "hfd", "aarch64_sve_fnmla", "aarch64_sve_fnmla_u">;
844-
defm SVNMLS_F : SInstZPZZZ<"svnmls", "hfd", "aarch64_sve_fnmls", "aarch64_sve_fnmls_u">;
845-
defm SVNMSB_F : SInstZPZZZ<"svnmsb", "hfd", "aarch64_sve_fnmsb", "aarch64_sve_fnmls_u", [ReverseMergeAnyAccOp]>;
838+
defm SVMAD_F : SInstZPZZZ<"svmad", "hfd", "aarch64_sve_fmad", "aarch64_sve_fmla_u", [IsStreamingCompatible, ReverseMergeAnyAccOp]>;
839+
defm SVMLA_F : SInstZPZZZ<"svmla", "hfd", "aarch64_sve_fmla", "aarch64_sve_fmla_u", [IsStreamingCompatible]>;
840+
defm SVMLS_F : SInstZPZZZ<"svmls", "hfd", "aarch64_sve_fmls", "aarch64_sve_fmls_u", [IsStreamingCompatible]>;
841+
defm SVMSB_F : SInstZPZZZ<"svmsb", "hfd", "aarch64_sve_fmsb", "aarch64_sve_fmls_u", [IsStreamingCompatible, ReverseMergeAnyAccOp]>;
842+
defm SVNMAD_F : SInstZPZZZ<"svnmad", "hfd", "aarch64_sve_fnmad", "aarch64_sve_fnmla_u", [IsStreamingCompatible, ReverseMergeAnyAccOp]>;
843+
defm SVNMLA_F : SInstZPZZZ<"svnmla", "hfd", "aarch64_sve_fnmla", "aarch64_sve_fnmla_u", [IsStreamingCompatible]>;
844+
defm SVNMLS_F : SInstZPZZZ<"svnmls", "hfd", "aarch64_sve_fnmls", "aarch64_sve_fnmls_u", [IsStreamingCompatible]>;
845+
defm SVNMSB_F : SInstZPZZZ<"svnmsb", "hfd", "aarch64_sve_fnmsb", "aarch64_sve_fnmls_u", [IsStreamingCompatible, ReverseMergeAnyAccOp]>;
846846

847847
def SVCADD_M : SInst<"svcadd[_{d}]", "dPddi", "hfd", MergeOp1, "aarch64_sve_fcadd", [IsStreamingCompatible], [ImmCheck<3, ImmCheckComplexRot90_270>]>;
848848
def SVCADD_X : SInst<"svcadd[_{d}]", "dPddi", "hfd", MergeAny, "aarch64_sve_fcadd", [IsStreamingCompatible], [ImmCheck<3, ImmCheckComplexRot90_270>]>;
@@ -881,11 +881,11 @@ def SVACLE : SInst<"svacle[_{d}]", "PPdd", "hfd", MergeNone, "aarch64_sve_facg
881881
def SVACLT : SInst<"svaclt[_{d}]", "PPdd", "hfd", MergeNone, "aarch64_sve_facgt", [ReverseCompare, IsStreamingCompatible]>;
882882
def SVCMPUO : SInst<"svcmpuo[_{d}]", "PPdd", "hfd", MergeNone, "aarch64_sve_fcmpuo", [IsStreamingCompatible]>;
883883

884-
def SVACGE_N : SInst<"svacge[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facge">;
885-
def SVACGT_N : SInst<"svacgt[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facgt">;
886-
def SVACLE_N : SInst<"svacle[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facge", [ReverseCompare]>;
887-
def SVACLT_N : SInst<"svaclt[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facgt", [ReverseCompare]>;
888-
def SVCMPUO_N : SInst<"svcmpuo[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_fcmpuo">;
884+
def SVACGE_N : SInst<"svacge[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facge", [IsStreamingCompatible]>;
885+
def SVACGT_N : SInst<"svacgt[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facgt", [IsStreamingCompatible]>;
886+
def SVACLE_N : SInst<"svacle[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facge", [ReverseCompare, IsStreamingCompatible]>;
887+
def SVACLT_N : SInst<"svaclt[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_facgt", [ReverseCompare, IsStreamingCompatible]>;
888+
def SVCMPUO_N : SInst<"svcmpuo[_n_{d}]", "PPda", "hfd", MergeNone, "aarch64_sve_fcmpuo", [IsStreamingCompatible]>;
889889

890890
def SVCMPEQ_F : SInst<"svcmpeq[_{d}]", "PPdd", "hfd", MergeNone, "aarch64_sve_fcmpeq", [IsStreamingCompatible]>;
891891
def SVCMPNE_F : SInst<"svcmpne[_{d}]", "PPdd", "hfd", MergeNone, "aarch64_sve_fcmpne", [IsStreamingCompatible]>;
@@ -1023,15 +1023,15 @@ def SVCOMPACT : SInst<"svcompact[_{d}]", "dPd", "ilUiUlfd", MergeNo
10231023
// splat of any possible lane. It is upto LLVM to pick a more efficient
10241024
// instruction such as DUP (indexed) if the lane index fits the range of the
10251025
// instruction's immediate.
1026-
def SVDUP_LANE : SInst<"svdup_lane[_{d}]", "ddL", "csilUcUsUiUlhfd", MergeNone, "aarch64_sve_tbl">;
1026+
def SVDUP_LANE : SInst<"svdup_lane[_{d}]", "ddL", "csilUcUsUiUlhfd", MergeNone, "aarch64_sve_tbl", [IsStreamingCompatible]>;
10271027
let TargetGuard = "sve,bf16" in {
10281028
def SVDUP_LANE_BF16 :
1029-
SInst<"svdup_lane[_{d}]", "ddL", "b", MergeNone, "aarch64_sve_tbl">;
1029+
SInst<"svdup_lane[_{d}]", "ddL", "b", MergeNone, "aarch64_sve_tbl", [IsStreamingCompatible]>;
10301030
}
10311031

1032-
def SVDUPQ_LANE : SInst<"svdupq_lane[_{d}]", "ddn", "csilUcUsUiUlhfd", MergeNone, "aarch64_sve_dupq_lane">;
1032+
def SVDUPQ_LANE : SInst<"svdupq_lane[_{d}]", "ddn", "csilUcUsUiUlhfd", MergeNone, "aarch64_sve_dupq_lane", [IsStreamingCompatible]>;
10331033
let TargetGuard = "sve,bf16" in {
1034-
def SVDUPQ_LANE_BF16 : SInst<"svdupq_lane[_{d}]", "ddn", "b", MergeNone, "aarch64_sve_dupq_lane">;
1034+
def SVDUPQ_LANE_BF16 : SInst<"svdupq_lane[_{d}]", "ddn", "b", MergeNone, "aarch64_sve_dupq_lane", [IsStreamingCompatible]>;
10351035
}
10361036
def SVEXT : SInst<"svext[_{d}]", "dddi", "csilUcUsUiUlhfd", MergeNone, "aarch64_sve_ext", [IsStreamingCompatible], [ImmCheck<2, ImmCheckExtract, 1>]>;
10371037
defm SVLASTA : SVEPerm<"svlasta[_{d}]", "sPd", "aarch64_sve_lasta">;
@@ -1109,11 +1109,11 @@ def SVPFALSE : SInst<"svpfalse[_b]", "Pv", "", MergeNone, "", [IsOverloadNone, I
11091109
def SVPTRUE_PAT : SInst<"svptrue_pat_{d}", "PI", "PcPsPiPl", MergeNone, "aarch64_sve_ptrue", [IsStreamingCompatible]>;
11101110
def SVPTRUE : SInst<"svptrue_{d}", "Pv", "PcPsPiPl", MergeNone, "aarch64_sve_ptrue", [IsAppendSVALL, IsStreamingCompatible]>;
11111111

1112-
def SVDUPQ_B8 : SInst<"svdupq[_n]_{d}", "Pssssssssssssssss", "Pc", MergeNone>;
1113-
def SVDUPQ_B16 : SInst<"svdupq[_n]_{d}", "Pssssssss", "Ps", MergeNone>;
1114-
def SVDUPQ_B32 : SInst<"svdupq[_n]_{d}", "Pssss", "Pi", MergeNone>;
1115-
def SVDUPQ_B64 : SInst<"svdupq[_n]_{d}", "Pss", "Pl", MergeNone>;
1116-
def SVDUP_N_B : SInst<"svdup[_n]_{d}", "Ps", "PcPsPiPl", MergeNone>;
1112+
def SVDUPQ_B8 : SInst<"svdupq[_n]_{d}", "Pssssssssssssssss", "Pc", MergeNone, "", [IsStreamingCompatible]>;
1113+
def SVDUPQ_B16 : SInst<"svdupq[_n]_{d}", "Pssssssss", "Ps", MergeNone, "", [IsStreamingCompatible]>;
1114+
def SVDUPQ_B32 : SInst<"svdupq[_n]_{d}", "Pssss", "Pi", MergeNone, "", [IsStreamingCompatible]>;
1115+
def SVDUPQ_B64 : SInst<"svdupq[_n]_{d}", "Pss", "Pl", MergeNone, "", [IsStreamingCompatible]>;
1116+
def SVDUP_N_B : SInst<"svdup[_n]_{d}", "Ps", "PcPsPiPl", MergeNone, "", [IsStreamingCompatible]>;
11171117

11181118

11191119
////////////////////////////////////////////////////////////////////////////////
@@ -1268,10 +1268,10 @@ def SVZIP2Q : SInst<"svzip2q[_{d}]", "ddd", "csilUcUsUiUlhfd", MergeNo
12681268
let TargetGuard = "sve,bf16,f64mm" in {
12691269
def SVTRN1Q_BF16 : SInst<"svtrn1q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_trn1q">;
12701270
def SVTRN2Q_BF16 : SInst<"svtrn2q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_trn2q">;
1271-
def SVUZP1Q_BF16 : SInst<"svuzp1q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_uzp1q">;
1272-
def SVUZP2Q_BF16 : SInst<"svuzp2q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_uzp2q">;
1273-
def SVZIP1Q_BF16 : SInst<"svzip1q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_zip1q">;
1274-
def SVZIP2Q_BF16 : SInst<"svzip2q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_zip2q">;
1271+
def SVUZP1Q_BF16 : SInst<"svuzp1q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_uzp1q", [IsStreamingCompatible]>;
1272+
def SVUZP2Q_BF16 : SInst<"svuzp2q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_uzp2q", [IsStreamingCompatible]>;
1273+
def SVZIP1Q_BF16 : SInst<"svzip1q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_zip1q", [IsStreamingCompatible]>;
1274+
def SVZIP2Q_BF16 : SInst<"svzip2q[_{d}]", "ddd", "b", MergeNone, "aarch64_sve_zip2q", [IsStreamingCompatible]>;
12751275
}
12761276

12771277
////////////////////////////////////////////////////////////////////////////////
@@ -1347,14 +1347,14 @@ def SVWHILEHS_U64 : SInst<"svwhilege_{d}[_{1}]", "Pnn", "PUcPUsPUiPUl", MergeNon
13471347
}
13481348

13491349
let TargetGuard = "sve2p1|sme2" in {
1350-
def SVWHILEGE_S64_X2 : SInst<"svwhilege_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilege_x2">;
1351-
def SVWHILEGT_S64_X2 : SInst<"svwhilegt_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilegt_x2">;
1352-
def SVWHILEHI_U64_X2 : SInst<"svwhilegt_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilehi_x2">;
1353-
def SVWHILEHS_U64_X2 : SInst<"svwhilege_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilehs_x2">;
1354-
def SVWHILELE_S64_X2 : SInst<"svwhilele_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilele_x2">;
1355-
def SVWHILELT_S64_X2 : SInst<"svwhilelt_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilelt_x2">;
1356-
def SVWHILELO_U64_X2 : SInst<"svwhilelt_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilelo_x2">;
1357-
def SVWHILELS_U64_X2 : SInst<"svwhilele_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilels_x2">;
1350+
def SVWHILEGE_S64_X2 : SInst<"svwhilege_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilege_x2", [IsStreamingOrSVE2p1]>;
1351+
def SVWHILEGT_S64_X2 : SInst<"svwhilegt_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilegt_x2", [IsStreamingOrSVE2p1]>;
1352+
def SVWHILEHI_U64_X2 : SInst<"svwhilegt_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilehi_x2", [IsStreamingOrSVE2p1]>;
1353+
def SVWHILEHS_U64_X2 : SInst<"svwhilege_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilehs_x2", [IsStreamingOrSVE2p1]>;
1354+
def SVWHILELE_S64_X2 : SInst<"svwhilele_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilele_x2", [IsStreamingOrSVE2p1]>;
1355+
def SVWHILELT_S64_X2 : SInst<"svwhilelt_{d}[_{1}]_x2", "2ll", "PcPsPiPl", MergeNone, "aarch64_sve_whilelt_x2", [IsStreamingOrSVE2p1]>;
1356+
def SVWHILELO_U64_X2 : SInst<"svwhilelt_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilelo_x2", [IsStreamingOrSVE2p1]>;
1357+
def SVWHILELS_U64_X2 : SInst<"svwhilele_{d}[_{1}]_x2", "2nn", "PcPsPiPl", MergeNone, "aarch64_sve_whilels_x2", [IsStreamingOrSVE2p1]>;
13581358

13591359
}
13601360

@@ -1831,8 +1831,8 @@ def SVPMULLB_PAIR : SInst<"svpmullb_pair[_{d}]", "ddd", "UcUi", Mer
18311831
def SVPMULLB_PAIR_N : SInst<"svpmullb_pair[_n_{d}]", "dda", "UcUi", MergeNone, "aarch64_sve_pmullb_pair", [IsStreamingCompatible]>;
18321832
def SVPMULLT : SInst<"svpmullt[_{d}]", "dhh", "UsUl", MergeNone, "", [IsStreamingCompatible]>;
18331833
def SVPMULLT_N : SInst<"svpmullt[_n_{d}]", "dhR", "UsUl", MergeNone, "", [IsStreamingCompatible]>;
1834-
def SVPMULLT_PAIR : SInst<"svpmullt_pair[_{d}]", "ddd", "UcUi", MergeNone, "aarch64_sve_pmullt_pair">;
1835-
def SVPMULLT_PAIR_N : SInst<"svpmullt_pair[_n_{d}]", "dda", "UcUi", MergeNone, "aarch64_sve_pmullt_pair">;
1834+
def SVPMULLT_PAIR : SInst<"svpmullt_pair[_{d}]", "ddd", "UcUi", MergeNone, "aarch64_sve_pmullt_pair", [IsStreamingCompatible]>;
1835+
def SVPMULLT_PAIR_N : SInst<"svpmullt_pair[_n_{d}]", "dda", "UcUi", MergeNone, "aarch64_sve_pmullt_pair", [IsStreamingCompatible]>;
18361836
}
18371837

18381838
////////////////////////////////////////////////////////////////////////////////

0 commit comments

Comments
 (0)