[clang] [Clang][NEON] Add neon target guard to intrinsics (PR #98624)

via cfe-commits cfe-commits at lists.llvm.org
Mon Jul 22 03:09:02 PDT 2024


https://github.com/Lukacma updated https://github.com/llvm/llvm-project/pull/98624

>From bbb844893d973f67ecc130595bfe4fe36b2cb707 Mon Sep 17 00:00:00 2001
From: Marian Lukac <Marian.Lukac at arm.com>
Date: Fri, 12 Jul 2024 12:17:29 +0000
Subject: [PATCH 1/2] [Clang][NEON] Add neon target guard to intrinsics

---
 clang/include/clang/Basic/arm_neon.td         | 64 +++++++++----------
 clang/include/clang/Basic/arm_neon_incl.td    |  2 +-
 .../aarch64-neon-without-target-feature.cpp   | 33 ++++++++++
 clang/utils/TableGen/NeonEmitter.cpp          |  6 +-
 4 files changed, 69 insertions(+), 36 deletions(-)
 create mode 100644 clang/test/Sema/aarch64-neon-without-target-feature.cpp

diff --git a/clang/include/clang/Basic/arm_neon.td b/clang/include/clang/Basic/arm_neon.td
index 6390ba3f9fe5e..3098fa67e6a51 100644
--- a/clang/include/clang/Basic/arm_neon.td
+++ b/clang/include/clang/Basic/arm_neon.td
@@ -289,7 +289,7 @@ def SPLATQ : WInst<"splat_laneq", ".(!Q)I",
                    "UcUsUicsilPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUlhdQhQdPlQPl"> {
   let isLaneQ = 1;
 }
-let TargetGuard = "bf16" in {
+let TargetGuard = "bf16,neon" in {
   def SPLAT_BF  : WInst<"splat_lane", ".(!q)I", "bQb">;
   def SPLATQ_BF : WInst<"splat_laneq", ".(!Q)I", "bQb"> {
     let isLaneQ = 1;
@@ -323,7 +323,7 @@ def VMLSL    : SOpInst<"vmlsl", "(>Q)(>Q)..", "csiUcUsUi", OP_MLSL>;
 def VQDMULH  : SInst<"vqdmulh", "...", "siQsQi">;
 def VQRDMULH : SInst<"vqrdmulh", "...", "siQsQi">;
 
-let TargetGuard = "v8.1a" in {
+let TargetGuard = "v8.1a,neon" in {
 def VQRDMLAH : SInst<"vqrdmlah", "....", "siQsQi">;
 def VQRDMLSH : SInst<"vqrdmlsh", "....", "siQsQi">;
 }
@@ -614,7 +614,7 @@ def A64_VQDMULH_LANE  : SInst<"vqdmulh_lane", "..(!q)I", "siQsQi">;
 def A64_VQRDMULH_LANE : SInst<"vqrdmulh_lane", "..(!q)I", "siQsQi">;
 }
 
-let TargetGuard = "v8.1a" in {
+let TargetGuard = "v8.1a,neon" in {
 def VQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "...qI", "siQsQi", OP_QRDMLAH_LN>;
 def VQRDMLSH_LANE : SOpInst<"vqrdmlsh_lane", "...qI", "siQsQi", OP_QRDMLSH_LN>;
 }
@@ -957,7 +957,7 @@ def VQDMLAL_HIGH : SOpInst<"vqdmlal_high", "(>Q)(>Q)QQ", "si", OP_QDMLALHi>;
 def VQDMLAL_HIGH_N : SOpInst<"vqdmlal_high_n", "(>Q)(>Q)Q1", "si", OP_QDMLALHi_N>;
 def VQDMLSL_HIGH : SOpInst<"vqdmlsl_high", "(>Q)(>Q)QQ", "si", OP_QDMLSLHi>;
 def VQDMLSL_HIGH_N : SOpInst<"vqdmlsl_high_n", "(>Q)(>Q)Q1", "si", OP_QDMLSLHi_N>;
-let TargetGuard = "aes" in {
+let TargetGuard = "aes,neon" in {
   def VMULL_P64    : SInst<"vmull", "(1>)11", "Pl">;
   def VMULL_HIGH_P64 : SOpInst<"vmull_high", "(1>)..", "HPl", OP_MULLHi_P64>;
 }
@@ -1091,7 +1091,7 @@ let isLaneQ = 1 in {
 def VQDMULH_LANEQ  : SInst<"vqdmulh_laneq", "..QI", "siQsQi">;
 def VQRDMULH_LANEQ : SInst<"vqrdmulh_laneq", "..QI", "siQsQi">;
 }
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.1a" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.1a,neon" in {
 def VQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "...QI", "siQsQi", OP_QRDMLAH_LN> {
   let isLaneQ = 1;
 }
@@ -1122,14 +1122,14 @@ def VEXT_A64 : WInst<"vext", "...I", "dQdPlQPl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Crypto
-let ArchGuard = "__ARM_ARCH >= 8", TargetGuard = "aes" in {
+let ArchGuard = "__ARM_ARCH >= 8", TargetGuard = "aes,neon" in {
 def AESE : SInst<"vaese", "...", "QUc">;
 def AESD : SInst<"vaesd", "...", "QUc">;
 def AESMC : SInst<"vaesmc", "..", "QUc">;
 def AESIMC : SInst<"vaesimc", "..", "QUc">;
 }
 
-let ArchGuard = "__ARM_ARCH >= 8", TargetGuard = "sha2" in {
+let ArchGuard = "__ARM_ARCH >= 8", TargetGuard = "sha2,neon" in {
 def SHA1H : SInst<"vsha1h", "11", "Ui">;
 def SHA1SU1 : SInst<"vsha1su1", "...", "QUi">;
 def SHA256SU0 : SInst<"vsha256su0", "...", "QUi">;
@@ -1143,7 +1143,7 @@ def SHA256H2 : SInst<"vsha256h2", "....", "QUi">;
 def SHA256SU1 : SInst<"vsha256su1", "....", "QUi">;
 }
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sha3" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sha3,neon" in {
 def BCAX : SInst<"vbcax", "....", "QUcQUsQUiQUlQcQsQiQl">;
 def EOR3 : SInst<"veor3", "....", "QUcQUsQUiQUlQcQsQiQl">;
 def RAX1 : SInst<"vrax1", "...", "QUl">;
@@ -1153,14 +1153,14 @@ def XAR :  SInst<"vxar", "...I", "QUl">;
 }
 }
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sha3" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sha3,neon" in {
 def SHA512SU0 : SInst<"vsha512su0", "...", "QUl">;
 def SHA512su1 : SInst<"vsha512su1", "....", "QUl">;
 def SHA512H : SInst<"vsha512h", "....", "QUl">;
 def SHA512H2 : SInst<"vsha512h2", "....", "QUl">;
 }
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sm4" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sm4,neon" in {
 def SM3SS1 : SInst<"vsm3ss1", "....", "QUi">;
 def SM3TT1A : SInst<"vsm3tt1a", "....I", "QUi">;
 def SM3TT1B : SInst<"vsm3tt1b", "....I", "QUi">;
@@ -1170,7 +1170,7 @@ def SM3PARTW1 : SInst<"vsm3partw1", "....", "QUi">;
 def SM3PARTW2 : SInst<"vsm3partw2", "....", "QUi">;
 }
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sm4" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "sm4,neon" in {
 def SM4E : SInst<"vsm4e", "...", "QUi">;
 def SM4EKEY : SInst<"vsm4ekey", "...", "QUi">;
 }
@@ -1227,7 +1227,7 @@ def FRINTZ_S64 : SInst<"vrnd", "..", "dQd">;
 def FRINTI_S64 : SInst<"vrndi", "..", "dQd">;
 }
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.5a" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.5a,neon" in {
 def FRINT32X_S32 : SInst<"vrnd32x", "..", "fQf">;
 def FRINT32Z_S32 : SInst<"vrnd32z", "..", "fQf">;
 def FRINT64X_S32 : SInst<"vrnd64x", "..", "fQf">;
@@ -1401,7 +1401,7 @@ def SCALAR_SQDMULH : SInst<"vqdmulh", "111", "SsSi">;
 // Scalar Integer Saturating Rounding Doubling Multiply Half High
 def SCALAR_SQRDMULH : SInst<"vqrdmulh", "111", "SsSi">;
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.1a" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.1a,neon" in {
 ////////////////////////////////////////////////////////////////////////////////
 // Signed Saturating Rounding Doubling Multiply Accumulate Returning High Half
 def SCALAR_SQRDMLAH : SInst<"vqrdmlah", "1111", "SsSi">;
@@ -1632,7 +1632,7 @@ def SCALAR_SQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "11QI", "SsSi", OP_SCALAR_
   let isLaneQ = 1;
 }
 
-let TargetGuard = "v8.1a" in {
+let TargetGuard = "v8.1a,neon" in {
 // Signed Saturating Rounding Doubling Multiply Accumulate Returning High Half
 def SCALAR_SQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "111.I", "SsSi", OP_SCALAR_QRDMLAH_LN>;
 def SCALAR_SQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "111QI", "SsSi", OP_SCALAR_QRDMLAH_LN> {
@@ -1654,7 +1654,7 @@ def SCALAR_VDUP_LANEQ : IInst<"vdup_laneq", "1QI", "ScSsSiSlSfSdSUcSUsSUiSUlSPcS
 } // ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)"
 
 // ARMv8.2-A FP16 vector intrinsics for A32/A64.
-let TargetGuard = "fullfp16" in {
+let TargetGuard = "fullfp16,neon" in {
 
   // ARMv8.2-A FP16 one-operand vector intrinsics.
 
@@ -1679,7 +1679,7 @@ let TargetGuard = "fullfp16" in {
   def VCVTP_U16    : SInst<"vcvtp_u16", "U.", "hQh">;
 
   // Vector rounding
-  let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_DIRECTED_ROUNDING)", TargetGuard = "fullfp16" in {
+  let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_DIRECTED_ROUNDING)", TargetGuard = "fullfp16,neon" in {
     def FRINTZH      : SInst<"vrnd",  "..", "hQh">;
     def FRINTNH      : SInst<"vrndn", "..", "hQh">;
     def FRINTAH      : SInst<"vrnda", "..", "hQh">;
@@ -1728,7 +1728,7 @@ let TargetGuard = "fullfp16" in {
   // Max/Min
   def VMAXH         : SInst<"vmax", "...", "hQh">;
   def VMINH         : SInst<"vmin", "...", "hQh">;
-  let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_NUMERIC_MAXMIN)", TargetGuard = "fullfp16" in {
+  let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_NUMERIC_MAXMIN)", TargetGuard = "fullfp16,neon" in {
     def FMAXNMH       : SInst<"vmaxnm", "...", "hQh">;
     def FMINNMH       : SInst<"vminnm", "...", "hQh">;
   }
@@ -1775,7 +1775,7 @@ def VEXTH      : WInst<"vext", "...I", "hQh">;
 def VREV64H    : WOpInst<"vrev64", "..", "hQh", OP_REV64>;
 
 // ARMv8.2-A FP16 vector intrinsics for A64 only.
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "fullfp16" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "fullfp16,neon" in {
 
   // Vector rounding
   def FRINTIH      : SInst<"vrndi", "..", "hQh">;
@@ -1872,11 +1872,11 @@ let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)" in {
 }
 
 // v8.2-A dot product instructions.
-let TargetGuard = "dotprod" in {
+let TargetGuard = "dotprod,neon" in {
   def DOT : SInst<"vdot", "..(<<)(<<)", "iQiUiQUi">;
   def DOT_LANE : SOpInst<"vdot_lane", "..(<<)(<<q)I", "iUiQiQUi", OP_DOT_LN>;
 }
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "dotprod" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "dotprod,neon" in {
   // Variants indexing into a 128-bit vector are A64 only.
   def UDOT_LANEQ : SOpInst<"vdot_laneq", "..(<<)(<<Q)I", "iUiQiQUi", OP_DOT_LNQ> {
     let isLaneQ = 1;
@@ -1884,7 +1884,7 @@ let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "d
 }
 
 // v8.2-A FP16 fused multiply-add long instructions.
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "fp16fml" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "fp16fml,neon" in {
   def VFMLAL_LOW  : SInst<"vfmlal_low",  ">>..", "hQh">;
   def VFMLSL_LOW  : SInst<"vfmlsl_low",  ">>..", "hQh">;
   def VFMLAL_HIGH : SInst<"vfmlal_high", ">>..", "hQh">;
@@ -1909,7 +1909,7 @@ let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "f
   }
 }
 
-let TargetGuard = "i8mm" in {
+let TargetGuard = "i8mm,neon" in {
   def VMMLA   : SInst<"vmmla", "..(<<)(<<)", "QUiQi">;
   def VUSMMLA : SInst<"vusmmla", "..(<<U)(<<)", "Qi">;
 
@@ -1926,7 +1926,7 @@ let TargetGuard = "i8mm" in {
   }
 }
 
-let TargetGuard = "bf16" in {
+let TargetGuard = "bf16,neon" in {
   def VDOT_BF : SInst<"vbfdot", "..BB", "fQf">;
   def VDOT_LANE_BF : SOpInst<"vbfdot_lane", "..B(Bq)I", "fQf", OP_BFDOT_LN>;
   def VDOT_LANEQ_BF : SOpInst<"vbfdot_laneq", "..B(BQ)I", "fQf", OP_BFDOT_LNQ> {
@@ -1970,7 +1970,7 @@ multiclass VCMLA_ROTS<string type, string lanety, string laneqty> {
 }
 
 // v8.3-A Vector complex addition intrinsics
-let TargetGuard = "v8.3a,fullfp16" in {
+let TargetGuard = "v8.3a,fullfp16,neon" in {
   def VCADD_ROT90_FP16   : SInst<"vcadd_rot90", "...", "h">;
   def VCADD_ROT270_FP16  : SInst<"vcadd_rot270", "...", "h">;
   def VCADDQ_ROT90_FP16  : SInst<"vcaddq_rot90", "QQQ", "h">;
@@ -1978,7 +1978,7 @@ let TargetGuard = "v8.3a,fullfp16" in {
 
   defm VCMLA_FP16  : VCMLA_ROTS<"h", "uint32x2_t", "uint32x4_t">;
 }
-let TargetGuard = "v8.3a" in {
+let TargetGuard = "v8.3a,neon" in {
   def VCADD_ROT90   : SInst<"vcadd_rot90", "...", "f">;
   def VCADD_ROT270  : SInst<"vcadd_rot270", "...", "f">;
   def VCADDQ_ROT90  : SInst<"vcaddq_rot90", "QQQ", "f">;
@@ -1986,7 +1986,7 @@ let TargetGuard = "v8.3a" in {
 
   defm VCMLA_F32        : VCMLA_ROTS<"f", "uint64x1_t", "uint64x2_t">;
 }
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.3a" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v8.3a,neon" in {
   def VCADDQ_ROT90_FP64  : SInst<"vcaddq_rot90", "QQQ", "d">;
   def VCADDQ_ROT270_FP64 : SInst<"vcaddq_rot270", "QQQ", "d">;
 
@@ -1994,7 +1994,7 @@ let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "v
 }
 
 // V8.2-A BFloat intrinsics
-let TargetGuard = "bf16" in {
+let TargetGuard = "bf16,neon" in {
   def VCREATE_BF : NoTestOpInst<"vcreate", ".(IU>)", "b", OP_CAST> {
     let BigEndianSafe = 1;
   }
@@ -2058,14 +2058,14 @@ let TargetGuard = "bf16" in {
   def SCALAR_CVT_F32_BF16 : SOpInst<"vcvtah_f32", "(1F>)(1!)", "b", OP_CVT_F32_BF16>;
 }
 
-let ArchGuard = "!defined(__aarch64__) && !defined(__arm64ec__)", TargetGuard = "bf16" in {
+let ArchGuard = "!defined(__aarch64__) && !defined(__arm64ec__)", TargetGuard = "bf16,neon" in {
   def VCVT_BF16_F32_A32_INTERNAL : WInst<"__a32_vcvt_bf16", "BQ", "f">;
   def VCVT_BF16_F32_A32 : SOpInst<"vcvt_bf16", "BQ", "f", OP_VCVT_BF16_F32_A32>;
   def VCVT_LOW_BF16_F32_A32 : SOpInst<"vcvt_low_bf16",  "BQ", "Qf", OP_VCVT_BF16_F32_LO_A32>;
   def VCVT_HIGH_BF16_F32_A32 : SOpInst<"vcvt_high_bf16", "BBQ", "Qf", OP_VCVT_BF16_F32_HI_A32>;
 }
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "bf16" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "bf16,neon" in {
   def VCVT_LOW_BF16_F32_A64_INTERNAL : WInst<"__a64_vcvtq_low_bf16", "BQ", "Hf">;
   def VCVT_LOW_BF16_F32_A64 : SOpInst<"vcvt_low_bf16", "BQ", "Qf", OP_VCVT_BF16_F32_LO_A64>;
   def VCVT_HIGH_BF16_F32_A64 : SInst<"vcvt_high_bf16", "BBQ", "Qf">;
@@ -2077,14 +2077,14 @@ let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "b
   def COPYQ_LANEQ_BF16 : IOpInst<"vcopy_laneq", "..I.I", "Qb", OP_COPY_LN>;
 }
 
-let ArchGuard = "!defined(__aarch64__) && !defined(__arm64ec__)", TargetGuard = "bf16" in {
+let ArchGuard = "!defined(__aarch64__) && !defined(__arm64ec__)", TargetGuard = "bf16,neon" in {
   let BigEndianSafe = 1 in {
     defm VREINTERPRET_BF : REINTERPRET_CROSS_TYPES<
         "csilUcUsUiUlhfPcPsPlQcQsQiQlQUcQUsQUiQUlQhQfQPcQPsQPl", "bQb">;
   }
 }
 
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "bf16" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "bf16,neon" in {
   let BigEndianSafe = 1 in {
     defm VVREINTERPRET_BF : REINTERPRET_CROSS_TYPES<
         "csilUcUsUiUlhfdPcPsPlQcQsQiQlQUcQUsQUiQUlQhQfQdQPcQPsQPlQPk", "bQb">;
@@ -2092,7 +2092,7 @@ let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "b
 }
 
 // v8.9a/v9.4a LRCPC3 intrinsics
-let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "rcpc3" in {
+let ArchGuard = "defined(__aarch64__) || defined(__arm64ec__)", TargetGuard = "rcpc3,neon" in {
   def VLDAP1_LANE : WInst<"vldap1_lane", ".(c*!).I", "QUlQlUlldQdPlQPl">;
   def VSTL1_LANE  : WInst<"vstl1_lane", "v*(.!)I", "QUlQlUlldQdPlQPl">;
 }
diff --git a/clang/include/clang/Basic/arm_neon_incl.td b/clang/include/clang/Basic/arm_neon_incl.td
index b8155c187d1bc..3b8015daee6d9 100644
--- a/clang/include/clang/Basic/arm_neon_incl.td
+++ b/clang/include/clang/Basic/arm_neon_incl.td
@@ -265,7 +265,7 @@ class Inst <string n, string p, string t, Operation o> {
   string Prototype = p;
   string Types = t;
   string ArchGuard = "";
-  string TargetGuard = "";
+  string TargetGuard = "neon";
 
   Operation Operation = o;
   bit BigEndianSafe = 0;
diff --git a/clang/test/Sema/aarch64-neon-without-target-feature.cpp b/clang/test/Sema/aarch64-neon-without-target-feature.cpp
new file mode 100644
index 0000000000000..f72dc0ce8df7e
--- /dev/null
+++ b/clang/test/Sema/aarch64-neon-without-target-feature.cpp
@@ -0,0 +1,33 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +dotprod  -target-feature +fullfp16 -target-feature +fp16fml -target-feature +i8mm -target-feature +bf16 -verify -emit-llvm -o - %s
+
+// This test is testing the diagnostics that Clang emits when compiling without '+neon'.
+
+#include <arm_neon.h>
+
+void undefined(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t v16i8, uint8x8_t v8i8, float32x2_t v2f32, float32x4_t v4f32, float16x4_t v4f16, float64x2_t v2f64, bfloat16x4_t v4bf16, __bf16 bf16, poly64_t poly64, poly64x2_t poly64x2) {
+  // dotprod
+  vdot_u32(v2i32, v8i8, v8i8); // expected-error {{always_inline function 'vdot_u32' requires target feature 'neon'}}
+  vdot_laneq_u32(v2i32, v8i8, v16i8, 1); // expected-error {{always_inline function 'vdot_u32' requires target feature 'neon'}} expected-error {{'__builtin_neon_splat_laneq_v' needs target feature neon}}
+  // fp16
+  vceqz_f16(v4f16); // expected-error {{always_inline function 'vceqz_f16' requires target feature 'neon'}}
+  vrnd_f16(v4f16); // expected-error {{always_inline function 'vrnd_f16' requires target feature 'neon'}}
+  vmaxnm_f16(v4f16, v4f16); // expected-error {{always_inline function 'vmaxnm_f16' requires target feature 'neon'}}
+  vrndi_f16(v4f16); // expected-error {{always_inline function 'vrndi_f16' requires target feature 'neon'}}
+  // fp16fml depends on fp-armv8
+  vfmlal_low_f16(v2f32, v4f16, v4f16); // expected-error {{always_inline function 'vfmlal_low_f16' requires target feature 'neon'}}
+  // i8mm
+  vmmlaq_s32(v4i32, v8i16, v8i16); // expected-error {{always_inline function 'vmmlaq_s32' requires target feature 'neon'}}
+  vusdot_laneq_s32(v2i32, v8i8, v8i16, 0); // expected-error {{always_inline function 'vusdot_s32' requires target feature 'neon'}} expected-error {{'__builtin_neon_splat_laneq_v' needs target feature neon}}
+  // bf16
+  vbfdot_f32(v2f32, v4bf16, v4bf16); // expected-error {{always_inline function 'vbfdot_f32' requires target feature 'neon'}}
+  vcreate_bf16(10);
+  vdup_lane_bf16(v4bf16, 2); // expected-error {{'__builtin_neon_splat_lane_bf16' needs target feature bf16,neon}}
+  vdup_n_bf16(bf16); // expected-error {{always_inline function 'vdup_n_bf16' requires target feature 'neon'}}
+  vld1_bf16(0); // expected-error {{'__builtin_neon_vld1_bf16' needs target feature bf16,neon}}
+  vcvt_f32_bf16(v4bf16); // expected-error {{always_inline function 'vcvt_f32_bf16' requires target feature 'neon'}}
+  vcvt_bf16_f32(v4f32); // expected-error {{always_inline function 'vcvt_bf16_f32' requires target feature 'neon'}}
+  vmull_p64(poly64, poly64);  // expected-error {{always_inline function 'vmull_p64' requires target feature 'neon'}}
+  vmull_high_p64(poly64x2, poly64x2);  // expected-error {{always_inline function 'vmull_high_p64' requires target feature 'neon'}}
+  vtrn1_s8(v8i8, v8i8); // expected-error {{always_inline function 'vtrn1_s8' requires target feature 'neon'}}
+
+}
diff --git a/clang/utils/TableGen/NeonEmitter.cpp b/clang/utils/TableGen/NeonEmitter.cpp
index 626031d38cf00..30fbb8c5d65e5 100644
--- a/clang/utils/TableGen/NeonEmitter.cpp
+++ b/clang/utils/TableGen/NeonEmitter.cpp
@@ -952,7 +952,7 @@ std::string Intrinsic::getInstTypeCode(Type T, ClassKind CK) const {
   char typeCode = '\0';
   bool printNumber = true;
 
-  if (CK == ClassB && TargetGuard == "")
+  if (CK == ClassB && TargetGuard == "neon")
     return "";
 
   if (T.isBFloat16())
@@ -976,7 +976,7 @@ std::string Intrinsic::getInstTypeCode(Type T, ClassKind CK) const {
       break;
     }
   }
-  if (CK == ClassB && TargetGuard == "") {
+  if (CK == ClassB && TargetGuard == "neon") {
     typeCode = '\0';
   }
 
@@ -1078,7 +1078,7 @@ std::string Intrinsic::mangleName(std::string Name, ClassKind LocalCK) const {
     S += "_" + getInstTypeCode(InBaseType, LocalCK);
   }
 
-  if (LocalCK == ClassB && TargetGuard == "")
+  if (LocalCK == ClassB && TargetGuard == "neon")
     S += "_v";
 
   // Insert a 'q' before the first '_' character so that it ends up before

>From 4ca19a5199b41344134b76275407c8c64143f1df Mon Sep 17 00:00:00 2001
From: Marian Lukac <Marian.Lukac at arm.com>
Date: Wed, 17 Jul 2024 16:34:10 +0000
Subject: [PATCH 2/2] Add tests for vbslq and vqabsq

---
 clang/test/Sema/aarch64-neon-without-target-feature.cpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/clang/test/Sema/aarch64-neon-without-target-feature.cpp b/clang/test/Sema/aarch64-neon-without-target-feature.cpp
index f72dc0ce8df7e..64f76a7900faf 100644
--- a/clang/test/Sema/aarch64-neon-without-target-feature.cpp
+++ b/clang/test/Sema/aarch64-neon-without-target-feature.cpp
@@ -29,5 +29,6 @@ void undefined(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t
   vmull_p64(poly64, poly64);  // expected-error {{always_inline function 'vmull_p64' requires target feature 'neon'}}
   vmull_high_p64(poly64x2, poly64x2);  // expected-error {{always_inline function 'vmull_high_p64' requires target feature 'neon'}}
   vtrn1_s8(v8i8, v8i8); // expected-error {{always_inline function 'vtrn1_s8' requires target feature 'neon'}}
-
+  vqabsq_s16(v8i16); // expected-error {{always_inline function 'vqabsq_s16' requires target feature 'neon'}}
+  vbslq_s16(v8i16, v8i16, v8i16);// expected-error {{always_inline function 'vbslq_s16' requires target feature 'neon'}}
 }



More information about the cfe-commits mailing list