Intrinsic<[llvm_ptr_ty], [], [IntrNoMem]>;
}
+//===----------------------------------------------------------------------===//
+// Saturating Arithmentic
+
+let TargetPrefix = "arm" in { // All intrinsics start with "llvm.arm.".
+ def int_arm_qadd : GCCBuiltin<"__builtin_arm_qadd">,
+ Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
+ [IntrNoMem, Commutative]>;
+ def int_arm_qsub : GCCBuiltin<"__builtin_arm_qsub">,
+ Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+ def int_arm_ssat : GCCBuiltin<"__builtin_arm_ssat">,
+ Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+ def int_arm_usat : GCCBuiltin<"__builtin_arm_usat">,
+ Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+}
+
+//===----------------------------------------------------------------------===//
+// VFP
+
+let TargetPrefix = "arm" in { // All intrinsics start with "llvm.arm.".
+ def int_arm_get_fpscr : GCCBuiltin<"__builtin_arm_get_fpscr">,
+ Intrinsic<[llvm_i32_ty], [], [IntrNoMem]>;
+ def int_arm_set_fpscr : GCCBuiltin<"__builtin_arm_set_fpscr">,
+ Intrinsic<[], [llvm_i32_ty], []>;
+ def int_arm_vcvtr : Intrinsic<[llvm_float_ty], [llvm_anyfloat_ty],
+ [IntrNoMem]>;
+ def int_arm_vcvtru : Intrinsic<[llvm_float_ty], [llvm_anyfloat_ty],
+ [IntrNoMem]>;
+}
+
//===----------------------------------------------------------------------===//
// Advanced SIMD (NEON)
// The following classes do not correspond directly to GCC builtins.
class Neon_1Arg_Intrinsic
- : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>], [IntrNoMem]>;
- class Neon_1Arg_Float_Intrinsic
- : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+ : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>], [IntrNoMem]>;
class Neon_1Arg_Narrow_Intrinsic
- : Intrinsic<[llvm_anyint_ty],
+ : Intrinsic<[llvm_anyvector_ty],
[LLVMExtendedElementVectorType<0>], [IntrNoMem]>;
- class Neon_1Arg_Long_Intrinsic
- : Intrinsic<[llvm_anyint_ty],
- [LLVMTruncatedElementVectorType<0>], [IntrNoMem]>;
class Neon_2Arg_Intrinsic
- : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
- [IntrNoMem]>;
- class Neon_2Arg_Float_Intrinsic
- : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
- [IntrNoMem]>;
- class Neon_2Arg_Vector_Intrinsic
: Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
[IntrNoMem]>;
class Neon_2Arg_Narrow_Intrinsic
- : Intrinsic<[llvm_anyint_ty],
+ : Intrinsic<[llvm_anyvector_ty],
[LLVMExtendedElementVectorType<0>,
LLVMExtendedElementVectorType<0>],
[IntrNoMem]>;
class Neon_2Arg_Long_Intrinsic
- : Intrinsic<[llvm_anyint_ty],
+ : Intrinsic<[llvm_anyvector_ty],
[LLVMTruncatedElementVectorType<0>,
LLVMTruncatedElementVectorType<0>],
[IntrNoMem]>;
- class Neon_2Arg_Wide_Intrinsic
- : Intrinsic<[llvm_anyint_ty],
- [LLVMMatchType<0>, LLVMTruncatedElementVectorType<0>],
- [IntrNoMem]>;
class Neon_3Arg_Intrinsic
- : Intrinsic<[llvm_anyint_ty],
+ : Intrinsic<[llvm_anyvector_ty],
[LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
[IntrNoMem]>;
class Neon_3Arg_Long_Intrinsic
- : Intrinsic<[llvm_anyint_ty],
+ : Intrinsic<[llvm_anyvector_ty],
[LLVMMatchType<0>,
LLVMTruncatedElementVectorType<0>,
LLVMTruncatedElementVectorType<0>],
[IntrNoMem]>;
- class Neon_2Result_Intrinsic
- : Intrinsic<[llvm_anyint_ty, LLVMMatchType<0>],
- [LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
- class Neon_2Result_Float_Intrinsic
- : Intrinsic<[llvm_anyfloat_ty, LLVMMatchType<0>],
- [LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
class Neon_CvtFxToFP_Intrinsic
: Intrinsic<[llvm_anyfloat_ty], [llvm_anyint_ty, llvm_i32_ty], [IntrNoMem]>;
class Neon_CvtFPToFx_Intrinsic
def int_arm_neon_vqaddu : Neon_2Arg_Intrinsic;
def int_arm_neon_vaddhn : Neon_2Arg_Narrow_Intrinsic;
def int_arm_neon_vraddhn : Neon_2Arg_Narrow_Intrinsic;
- def int_arm_neon_vaddls : Neon_2Arg_Long_Intrinsic;
- def int_arm_neon_vaddlu : Neon_2Arg_Long_Intrinsic;
- def int_arm_neon_vaddws : Neon_2Arg_Wide_Intrinsic;
- def int_arm_neon_vaddwu : Neon_2Arg_Wide_Intrinsic;
// Vector Multiply.
def int_arm_neon_vmulp : Neon_2Arg_Intrinsic;
def int_arm_neon_vqdmulh : Neon_2Arg_Intrinsic;
def int_arm_neon_vqrdmulh : Neon_2Arg_Intrinsic;
- def int_arm_neon_vmulls : Neon_2Arg_Long_Intrinsic;
- def int_arm_neon_vmullu : Neon_2Arg_Long_Intrinsic;
def int_arm_neon_vmullp : Neon_2Arg_Long_Intrinsic;
def int_arm_neon_vqdmull : Neon_2Arg_Long_Intrinsic;
-
- // Vector Multiply and Accumulate/Subtract.
- def int_arm_neon_vmlals : Neon_3Arg_Long_Intrinsic;
- def int_arm_neon_vmlalu : Neon_3Arg_Long_Intrinsic;
- def int_arm_neon_vmlsls : Neon_3Arg_Long_Intrinsic;
- def int_arm_neon_vmlslu : Neon_3Arg_Long_Intrinsic;
def int_arm_neon_vqdmlal : Neon_3Arg_Long_Intrinsic;
def int_arm_neon_vqdmlsl : Neon_3Arg_Long_Intrinsic;
// Vector Maximum.
def int_arm_neon_vmaxs : Neon_2Arg_Intrinsic;
def int_arm_neon_vmaxu : Neon_2Arg_Intrinsic;
- def int_arm_neon_vmaxf : Neon_2Arg_Float_Intrinsic;
// Vector Minimum.
def int_arm_neon_vmins : Neon_2Arg_Intrinsic;
def int_arm_neon_vminu : Neon_2Arg_Intrinsic;
- def int_arm_neon_vminf : Neon_2Arg_Float_Intrinsic;
// Vector Reciprocal Step.
- def int_arm_neon_vrecps : Neon_2Arg_Float_Intrinsic;
+ def int_arm_neon_vrecps : Neon_2Arg_Intrinsic;
// Vector Reciprocal Square Root Step.
- def int_arm_neon_vrsqrts : Neon_2Arg_Float_Intrinsic;
+ def int_arm_neon_vrsqrts : Neon_2Arg_Intrinsic;
}
// Vector Subtract.
def int_arm_neon_vqsubu : Neon_2Arg_Intrinsic;
def int_arm_neon_vsubhn : Neon_2Arg_Narrow_Intrinsic;
def int_arm_neon_vrsubhn : Neon_2Arg_Narrow_Intrinsic;
-def int_arm_neon_vsubls : Neon_2Arg_Long_Intrinsic;
-def int_arm_neon_vsublu : Neon_2Arg_Long_Intrinsic;
-def int_arm_neon_vsubws : Neon_2Arg_Wide_Intrinsic;
-def int_arm_neon_vsubwu : Neon_2Arg_Wide_Intrinsic;
// Vector Absolute Compare.
let TargetPrefix = "arm" in {
// Vector Absolute Differences.
def int_arm_neon_vabds : Neon_2Arg_Intrinsic;
def int_arm_neon_vabdu : Neon_2Arg_Intrinsic;
-def int_arm_neon_vabdf : Neon_2Arg_Float_Intrinsic;
-def int_arm_neon_vabdls : Neon_2Arg_Long_Intrinsic;
-def int_arm_neon_vabdlu : Neon_2Arg_Long_Intrinsic;
-
-// Vector Absolute Difference and Accumulate.
-def int_arm_neon_vabas : Neon_3Arg_Intrinsic;
-def int_arm_neon_vabau : Neon_3Arg_Intrinsic;
-def int_arm_neon_vabals : Neon_3Arg_Long_Intrinsic;
-def int_arm_neon_vabalu : Neon_3Arg_Long_Intrinsic;
// Vector Pairwise Add.
-def int_arm_neon_vpadd : Neon_2Arg_Vector_Intrinsic;
+def int_arm_neon_vpadd : Neon_2Arg_Intrinsic;
// Vector Pairwise Add Long.
// Note: This is different than the other "long" NEON intrinsics because
// the result vector has half as many elements as the source vector.
// The source and destination vector types must be specified separately.
let TargetPrefix = "arm" in {
- def int_arm_neon_vpaddls : Intrinsic<[llvm_anyint_ty], [llvm_anyint_ty],
+ def int_arm_neon_vpaddls : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty],
[IntrNoMem]>;
- def int_arm_neon_vpaddlu : Intrinsic<[llvm_anyint_ty], [llvm_anyint_ty],
+ def int_arm_neon_vpaddlu : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty],
[IntrNoMem]>;
}
// Note: This is similar to vpaddl but the destination vector also appears
// as the first argument.
let TargetPrefix = "arm" in {
- def int_arm_neon_vpadals : Intrinsic<[llvm_anyint_ty],
- [LLVMMatchType<0>, llvm_anyint_ty],
+ def int_arm_neon_vpadals : Intrinsic<[llvm_anyvector_ty],
+ [LLVMMatchType<0>, llvm_anyvector_ty],
[IntrNoMem]>;
- def int_arm_neon_vpadalu : Intrinsic<[llvm_anyint_ty],
- [LLVMMatchType<0>, llvm_anyint_ty],
+ def int_arm_neon_vpadalu : Intrinsic<[llvm_anyvector_ty],
+ [LLVMMatchType<0>, llvm_anyvector_ty],
[IntrNoMem]>;
}
// Vector Pairwise Maximum and Minimum.
def int_arm_neon_vpmaxs : Neon_2Arg_Intrinsic;
def int_arm_neon_vpmaxu : Neon_2Arg_Intrinsic;
-def int_arm_neon_vpmaxf : Neon_2Arg_Float_Intrinsic;
def int_arm_neon_vpmins : Neon_2Arg_Intrinsic;
def int_arm_neon_vpminu : Neon_2Arg_Intrinsic;
-def int_arm_neon_vpminf : Neon_2Arg_Float_Intrinsic;
// Vector Shifts:
//
// Vector Absolute Value and Saturating Absolute Value.
def int_arm_neon_vabs : Neon_1Arg_Intrinsic;
-def int_arm_neon_vabsf : Neon_1Arg_Float_Intrinsic;
def int_arm_neon_vqabs : Neon_1Arg_Intrinsic;
// Vector Saturating Negate.
// Vector Reciprocal Estimate.
def int_arm_neon_vrecpe : Neon_1Arg_Intrinsic;
-def int_arm_neon_vrecpef : Neon_1Arg_Float_Intrinsic;
// Vector Reciprocal Square Root Estimate.
def int_arm_neon_vrsqrte : Neon_1Arg_Intrinsic;
-def int_arm_neon_vrsqrtef : Neon_1Arg_Float_Intrinsic;
// Vector Conversions Between Floating-point and Fixed-point.
def int_arm_neon_vcvtfp2fxs : Neon_CvtFPToFx_Intrinsic;
def int_arm_neon_vcvtfxs2fp : Neon_CvtFxToFP_Intrinsic;
def int_arm_neon_vcvtfxu2fp : Neon_CvtFxToFP_Intrinsic;
-// Narrowing and Lengthening Vector Moves.
-def int_arm_neon_vmovn : Neon_1Arg_Narrow_Intrinsic;
+// Narrowing Saturating Vector Moves.
def int_arm_neon_vqmovns : Neon_1Arg_Narrow_Intrinsic;
def int_arm_neon_vqmovnu : Neon_1Arg_Narrow_Intrinsic;
def int_arm_neon_vqmovnsu : Neon_1Arg_Narrow_Intrinsic;
-def int_arm_neon_vmovls : Neon_1Arg_Long_Intrinsic;
-def int_arm_neon_vmovlu : Neon_1Arg_Long_Intrinsic;
// Vector Table Lookup.
+// The first 1-4 arguments are the table.
def int_arm_neon_vtbl1 : Neon_Tbl2Arg_Intrinsic;
def int_arm_neon_vtbl2 : Neon_Tbl3Arg_Intrinsic;
def int_arm_neon_vtbl3 : Neon_Tbl4Arg_Intrinsic;
def int_arm_neon_vtbl4 : Neon_Tbl5Arg_Intrinsic;
// Vector Table Extension.
+// Some elements of the destination vector may not be updated, so the original
+// value of that vector is passed as the first argument. The next 1-4
+// arguments after that are the table.
def int_arm_neon_vtbx1 : Neon_Tbl3Arg_Intrinsic;
def int_arm_neon_vtbx2 : Neon_Tbl4Arg_Intrinsic;
def int_arm_neon_vtbx3 : Neon_Tbl5Arg_Intrinsic;
def int_arm_neon_vtbx4 : Neon_Tbl6Arg_Intrinsic;
-// Vector Transpose.
-def int_arm_neon_vtrni : Neon_2Result_Intrinsic;
-def int_arm_neon_vtrnf : Neon_2Result_Float_Intrinsic;
-
-// Vector Interleave (vzip).
-def int_arm_neon_vzipi : Neon_2Result_Intrinsic;
-def int_arm_neon_vzipf : Neon_2Result_Float_Intrinsic;
-
-// Vector Deinterleave (vuzp).
-def int_arm_neon_vuzpi : Neon_2Result_Intrinsic;
-def int_arm_neon_vuzpf : Neon_2Result_Float_Intrinsic;
-
let TargetPrefix = "arm" in {
// De-interleaving vector loads from N-element structures.
- def int_arm_neon_vld1i : Intrinsic<[llvm_anyint_ty],
- [llvm_ptr_ty], [IntrReadArgMem]>;
- def int_arm_neon_vld1f : Intrinsic<[llvm_anyfloat_ty],
- [llvm_ptr_ty], [IntrReadArgMem]>;
- def int_arm_neon_vld2i : Intrinsic<[llvm_anyint_ty, LLVMMatchType<0>],
- [llvm_ptr_ty], [IntrReadArgMem]>;
- def int_arm_neon_vld2f : Intrinsic<[llvm_anyfloat_ty, LLVMMatchType<0>],
- [llvm_ptr_ty], [IntrReadArgMem]>;
- def int_arm_neon_vld3i : Intrinsic<[llvm_anyint_ty, LLVMMatchType<0>,
- LLVMMatchType<0>],
- [llvm_ptr_ty], [IntrReadArgMem]>;
- def int_arm_neon_vld3f : Intrinsic<[llvm_anyfloat_ty, LLVMMatchType<0>,
- LLVMMatchType<0>],
- [llvm_ptr_ty], [IntrReadArgMem]>;
- def int_arm_neon_vld4i : Intrinsic<[llvm_anyint_ty, LLVMMatchType<0>,
- LLVMMatchType<0>, LLVMMatchType<0>],
- [llvm_ptr_ty], [IntrReadArgMem]>;
- def int_arm_neon_vld4f : Intrinsic<[llvm_anyfloat_ty, LLVMMatchType<0>,
- LLVMMatchType<0>, LLVMMatchType<0>],
- [llvm_ptr_ty], [IntrReadArgMem]>;
+ // Source operands are the address and alignment.
+ def int_arm_neon_vld1 : Intrinsic<[llvm_anyvector_ty],
+ [llvm_ptr_ty, llvm_i32_ty],
+ [IntrReadArgMem]>;
+ def int_arm_neon_vld2 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>],
+ [llvm_ptr_ty, llvm_i32_ty],
+ [IntrReadArgMem]>;
+ def int_arm_neon_vld3 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
+ LLVMMatchType<0>],
+ [llvm_ptr_ty, llvm_i32_ty],
+ [IntrReadArgMem]>;
+ def int_arm_neon_vld4 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
+ LLVMMatchType<0>, LLVMMatchType<0>],
+ [llvm_ptr_ty, llvm_i32_ty],
+ [IntrReadArgMem]>;
+
+ // Vector load N-element structure to one lane.
+ // Source operands are: the address, the N input vectors (since only one
+ // lane is assigned), the lane number, and the alignment.
+ def int_arm_neon_vld2lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>],
+ [llvm_ptr_ty, LLVMMatchType<0>,
+ LLVMMatchType<0>, llvm_i32_ty,
+ llvm_i32_ty], [IntrReadArgMem]>;
+ def int_arm_neon_vld3lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
+ LLVMMatchType<0>],
+ [llvm_ptr_ty, LLVMMatchType<0>,
+ LLVMMatchType<0>, LLVMMatchType<0>,
+ llvm_i32_ty, llvm_i32_ty],
+ [IntrReadArgMem]>;
+ def int_arm_neon_vld4lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
+ LLVMMatchType<0>, LLVMMatchType<0>],
+ [llvm_ptr_ty, LLVMMatchType<0>,
+ LLVMMatchType<0>, LLVMMatchType<0>,
+ LLVMMatchType<0>, llvm_i32_ty,
+ llvm_i32_ty], [IntrReadArgMem]>;
// Interleaving vector stores from N-element structures.
- def int_arm_neon_vst1i : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyint_ty],
- [IntrWriteArgMem]>;
- def int_arm_neon_vst1f : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyfloat_ty],
- [IntrWriteArgMem]>;
- def int_arm_neon_vst2i : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyint_ty,
- LLVMMatchType<0>], [IntrWriteArgMem]>;
- def int_arm_neon_vst2f : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyfloat_ty,
- LLVMMatchType<0>], [IntrWriteArgMem]>;
- def int_arm_neon_vst3i : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyint_ty,
- LLVMMatchType<0>, LLVMMatchType<0>],
- [IntrWriteArgMem]>;
- def int_arm_neon_vst3f : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyfloat_ty,
- LLVMMatchType<0>, LLVMMatchType<0>],
- [IntrWriteArgMem]>;
- def int_arm_neon_vst4i : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyint_ty,
- LLVMMatchType<0>, LLVMMatchType<0>,
- LLVMMatchType<0>], [IntrWriteArgMem]>;
- def int_arm_neon_vst4f : Intrinsic<[llvm_void_ty],
- [llvm_ptr_ty, llvm_anyfloat_ty,
- LLVMMatchType<0>, LLVMMatchType<0>,
- LLVMMatchType<0>], [IntrWriteArgMem]>;
+ // Source operands are: the address, the N vectors, and the alignment.
+ def int_arm_neon_vst1 : Intrinsic<[],
+ [llvm_ptr_ty, llvm_anyvector_ty,
+ llvm_i32_ty], [IntrReadWriteArgMem]>;
+ def int_arm_neon_vst2 : Intrinsic<[],
+ [llvm_ptr_ty, llvm_anyvector_ty,
+ LLVMMatchType<0>, llvm_i32_ty],
+ [IntrReadWriteArgMem]>;
+ def int_arm_neon_vst3 : Intrinsic<[],
+ [llvm_ptr_ty, llvm_anyvector_ty,
+ LLVMMatchType<0>, LLVMMatchType<0>,
+ llvm_i32_ty], [IntrReadWriteArgMem]>;
+ def int_arm_neon_vst4 : Intrinsic<[],
+ [llvm_ptr_ty, llvm_anyvector_ty,
+ LLVMMatchType<0>, LLVMMatchType<0>,
+ LLVMMatchType<0>, llvm_i32_ty],
+ [IntrReadWriteArgMem]>;
+
+ // Vector store N-element structure from one lane.
+ // Source operands are: the address, the N vectors, the lane number, and
+ // the alignment.
+ def int_arm_neon_vst2lane : Intrinsic<[],
+ [llvm_ptr_ty, llvm_anyvector_ty,
+ LLVMMatchType<0>, llvm_i32_ty,
+ llvm_i32_ty], [IntrReadWriteArgMem]>;
+ def int_arm_neon_vst3lane : Intrinsic<[],
+ [llvm_ptr_ty, llvm_anyvector_ty,
+ LLVMMatchType<0>, LLVMMatchType<0>,
+ llvm_i32_ty, llvm_i32_ty],
+ [IntrReadWriteArgMem]>;
+ def int_arm_neon_vst4lane : Intrinsic<[],
+ [llvm_ptr_ty, llvm_anyvector_ty,
+ LLVMMatchType<0>, LLVMMatchType<0>,
+ LLVMMatchType<0>, llvm_i32_ty,
+ llvm_i32_ty], [IntrReadWriteArgMem]>;
}