diff options
Diffstat (limited to 'include/llvm/IR/IntrinsicsAArch64.td')
| -rw-r--r-- | include/llvm/IR/IntrinsicsAArch64.td | 201 |
1 files changed, 110 insertions, 91 deletions
diff --git a/include/llvm/IR/IntrinsicsAArch64.td b/include/llvm/IR/IntrinsicsAArch64.td index 68af8c1..61c0e5d 100644 --- a/include/llvm/IR/IntrinsicsAArch64.td +++ b/include/llvm/IR/IntrinsicsAArch64.td @@ -36,27 +36,11 @@ def int_aarch64_neon_xtn : // Vector floating-point convert def int_aarch64_neon_frintn : Neon_1Arg_Intrinsic; def int_aarch64_neon_fsqrt : Neon_1Arg_Intrinsic; -def int_aarch64_neon_fcvtxn : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtns : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtnu : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtps : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtpu : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtms : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtmu : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtas : +def int_aarch64_neon_vcvtxn : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtau : - Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtzs : +def int_aarch64_neon_vcvtzs : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -def int_aarch64_neon_fcvtzu : +def int_aarch64_neon_vcvtzu : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; // Vector maxNum (Floating Point) @@ -84,7 +68,7 @@ class Neon_N3V_Intrinsic [IntrNoMem]>; class Neon_N2V_Narrow_Intrinsic : Intrinsic<[llvm_anyvector_ty], - [LLVMExtendedElementVectorType<0>, llvm_i32_ty], + [LLVMExtendedType<0>, llvm_i32_ty], [IntrNoMem]>; // Vector rounding shift right by immediate (Signed) @@ -107,9 +91,6 @@ def int_aarch64_neon_vuqrshrn : Neon_N2V_Narrow_Intrinsic; class Neon_Across_Intrinsic : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>; -class Neon_2Arg_Across_Float_Intrinsic - : Intrinsic<[llvm_anyvector_ty], [llvm_v4f32_ty], [IntrNoMem]>; - def int_aarch64_neon_saddlv : Neon_Across_Intrinsic; def int_aarch64_neon_uaddlv : Neon_Across_Intrinsic; def int_aarch64_neon_smaxv : Neon_Across_Intrinsic; @@ -117,30 +98,34 @@ def int_aarch64_neon_umaxv : Neon_Across_Intrinsic; def int_aarch64_neon_sminv : Neon_Across_Intrinsic; def int_aarch64_neon_uminv : Neon_Across_Intrinsic; def int_aarch64_neon_vaddv : Neon_Across_Intrinsic; -def int_aarch64_neon_vmaxv : Neon_Across_Intrinsic; -def int_aarch64_neon_vminv : Neon_Across_Intrinsic; -def int_aarch64_neon_vmaxnmv : Neon_Across_Intrinsic; -def int_aarch64_neon_vminnmv : Neon_Across_Intrinsic; +def int_aarch64_neon_vmaxv : + Intrinsic<[llvm_float_ty], [llvm_v4f32_ty], [IntrNoMem]>; +def int_aarch64_neon_vminv : + Intrinsic<[llvm_float_ty], [llvm_v4f32_ty], [IntrNoMem]>; +def int_aarch64_neon_vmaxnmv : + Intrinsic<[llvm_float_ty], [llvm_v4f32_ty], [IntrNoMem]>; +def int_aarch64_neon_vminnmv : + Intrinsic<[llvm_float_ty], [llvm_v4f32_ty], [IntrNoMem]>; // Vector Table Lookup. def int_aarch64_neon_vtbl1 : Intrinsic<[llvm_anyvector_ty], - [llvm_anyvector_ty, LLVMMatchType<0>], [IntrNoMem]>; + [llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>; def int_aarch64_neon_vtbl2 : Intrinsic<[llvm_anyvector_ty], - [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<0>], + [llvm_v16i8_ty, llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>; def int_aarch64_neon_vtbl3 : Intrinsic<[llvm_anyvector_ty], - [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<1>, + [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>; def int_aarch64_neon_vtbl4 : Intrinsic<[llvm_anyvector_ty], - [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<1>, - LLVMMatchType<1>, LLVMMatchType<0>], [IntrNoMem]>; + [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty, + llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>; // Vector Table Extension. // Some elements of the destination vector may not be updated, so the original @@ -148,23 +133,23 @@ def int_aarch64_neon_vtbl4 : // arguments after that are the table. def int_aarch64_neon_vtbx1 : Intrinsic<[llvm_anyvector_ty], - [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<0>], + [LLVMMatchType<0>, llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>; def int_aarch64_neon_vtbx2 : Intrinsic<[llvm_anyvector_ty], - [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>, - LLVMMatchType<0>], [IntrNoMem]>; + [LLVMMatchType<0>, llvm_v16i8_ty, llvm_v16i8_ty, + LLVMMatchType<0>], [IntrNoMem]>; def int_aarch64_neon_vtbx3 : Intrinsic<[llvm_anyvector_ty], - [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>, - LLVMMatchType<1>, LLVMMatchType<0>], [IntrNoMem]>; + [LLVMMatchType<0>, llvm_v16i8_ty, llvm_v16i8_ty, + llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>; def int_aarch64_neon_vtbx4 : Intrinsic<[llvm_anyvector_ty], - [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>, - LLVMMatchType<1>, LLVMMatchType<1>, LLVMMatchType<0>], + [LLVMMatchType<0>, llvm_v16i8_ty, llvm_v16i8_ty, + llvm_v16i8_ty, llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>; // Vector Load/store @@ -233,74 +218,118 @@ def int_aarch64_neon_vqrshlu : Neon_2Arg_Intrinsic; def int_aarch64_neon_vpadd : Intrinsic<[llvm_v1i64_ty], [llvm_v2i64_ty],[IntrNoMem]>; def int_aarch64_neon_vpfadd : - Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>; -def int_aarch64_neon_vpfaddq : - Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>; + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>; // Scalar Reduce Pairwise Floating Point Max/Min. def int_aarch64_neon_vpmax : - Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>; -def int_aarch64_neon_vpmaxq : - Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>; + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>; def int_aarch64_neon_vpmin : - Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>; -def int_aarch64_neon_vpminq : - Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>; + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>; // Scalar Reduce Pairwise Floating Point Maxnm/Minnm. def int_aarch64_neon_vpfmaxnm : - Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>; -def int_aarch64_neon_vpfmaxnmq : - Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>; + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>; def int_aarch64_neon_vpfminnm : - Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>; -def int_aarch64_neon_vpfminnmq : - Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>; + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>; // Scalar Signed Integer Convert To Floating-point -def int_aarch64_neon_vcvtf32_s32 : - Intrinsic<[llvm_float_ty], [llvm_v1i32_ty], [IntrNoMem]>; -def int_aarch64_neon_vcvtf64_s64 : - Intrinsic<[llvm_double_ty], [llvm_v1i64_ty], [IntrNoMem]>; +def int_aarch64_neon_vcvtint2fps : + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>; // Scalar Unsigned Integer Convert To Floating-point -def int_aarch64_neon_vcvtf32_u32 : - Intrinsic<[llvm_float_ty], [llvm_v1i32_ty], [IntrNoMem]>; -def int_aarch64_neon_vcvtf64_u64 : - Intrinsic<[llvm_double_ty], [llvm_v1i64_ty], [IntrNoMem]>; +def int_aarch64_neon_vcvtint2fpu : + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>; + +// Scalar Floating-point Convert +def int_aarch64_neon_fcvtxn : + Intrinsic<[llvm_float_ty], [llvm_double_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtns : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtnu : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtps : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtpu : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtms : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtmu : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtas : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtau : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtzs : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; +def int_aarch64_neon_fcvtzu : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>; + +// Scalar Floating-point Reciprocal Estimate. +def int_aarch64_neon_vrecpe : + Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>; // Scalar Floating-point Reciprocal Exponent -def int_aarch64_neon_vrecpx : Neon_1Arg_Intrinsic; +def int_aarch64_neon_vrecpx : + Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>; -class Neon_Cmp_Intrinsic - : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty, llvm_anyvector_ty], - [IntrNoMem]>; +// Scalar Floating-point Reciprocal Square Root Estimate +def int_aarch64_neon_vrsqrte : + Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>; + +// Scalar Floating-point Reciprocal Step +def int_aarch64_neon_vrecps : + Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>], + [IntrNoMem]>; + +// Scalar Floating-point Reciprocal Square Root Step +def int_aarch64_neon_vrsqrts : + Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>], + [IntrNoMem]>; + +// Compare with vector operands. +class Neon_Cmp_Intrinsic : + Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty, llvm_anyvector_ty], + [IntrNoMem]>; + +// Floating-point compare with scalar operands. +class Neon_Float_Cmp_Intrinsic : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty, llvm_anyfloat_ty], + [IntrNoMem]>; // Scalar Compare Equal def int_aarch64_neon_vceq : Neon_Cmp_Intrinsic; +def int_aarch64_neon_fceq : Neon_Float_Cmp_Intrinsic; // Scalar Compare Greater-Than or Equal def int_aarch64_neon_vcge : Neon_Cmp_Intrinsic; def int_aarch64_neon_vchs : Neon_Cmp_Intrinsic; +def int_aarch64_neon_fcge : Neon_Float_Cmp_Intrinsic; +def int_aarch64_neon_fchs : Neon_Float_Cmp_Intrinsic; // Scalar Compare Less-Than or Equal def int_aarch64_neon_vclez : Neon_Cmp_Intrinsic; +def int_aarch64_neon_fclez : Neon_Float_Cmp_Intrinsic; // Scalar Compare Less-Than def int_aarch64_neon_vcltz : Neon_Cmp_Intrinsic; +def int_aarch64_neon_fcltz : Neon_Float_Cmp_Intrinsic; // Scalar Compare Greater-Than def int_aarch64_neon_vcgt : Neon_Cmp_Intrinsic; def int_aarch64_neon_vchi : Neon_Cmp_Intrinsic; +def int_aarch64_neon_fcgt : Neon_Float_Cmp_Intrinsic; +def int_aarch64_neon_fchi : Neon_Float_Cmp_Intrinsic; // Scalar Compare Bitwise Test Bits def int_aarch64_neon_vtstd : Neon_Cmp_Intrinsic; // Scalar Floating-point Absolute Compare Greater Than Or Equal def int_aarch64_neon_vcage : Neon_Cmp_Intrinsic; - +def int_aarch64_neon_fcage : Neon_Float_Cmp_Intrinsic; + // Scalar Floating-point Absolute Compare Greater Than def int_aarch64_neon_vcagt : Neon_Cmp_Intrinsic; +def int_aarch64_neon_fcagt : Neon_Float_Cmp_Intrinsic; // Scalar Signed Saturating Accumulated of Unsigned Value def int_aarch64_neon_vuqadd : Neon_2Arg_Intrinsic; @@ -313,7 +342,9 @@ def int_aarch64_neon_vabs : Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty], [IntrNoMem]>; // Scalar Absolute Difference -def int_aarch64_neon_vabd : Neon_2Arg_Intrinsic; +def int_aarch64_neon_vabd : + Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>], + [IntrNoMem]>; // Scalar Negate Value def int_aarch64_neon_vneg : @@ -325,6 +356,9 @@ def int_aarch64_neon_vqdmlal : Neon_3Arg_Long_Intrinsic; // Signed Saturating Doubling Multiply-Subtract Long def int_aarch64_neon_vqdmlsl : Neon_3Arg_Long_Intrinsic; +def int_aarch64_neon_vmull_p64 : + Intrinsic<[llvm_v16i8_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>; + class Neon_2Arg_ShiftImm_Intrinsic : Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>; @@ -355,34 +389,19 @@ def int_aarch64_neon_vqshlu_n : Neon_N2V_Intrinsic; def int_aarch64_neon_vqshlus_n : Neon_N2V_Intrinsic; // Scalar Signed Fixed-point Convert To Floating-Point (Immediate) -def int_aarch64_neon_vcvtf32_n_s32 : - Intrinsic<[llvm_float_ty], [llvm_v1i32_ty, llvm_i32_ty], [IntrNoMem]>; -def int_aarch64_neon_vcvtf64_n_s64 : - Intrinsic<[llvm_double_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>; +def int_aarch64_neon_vcvtfxs2fp_n : + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty, llvm_i32_ty], [IntrNoMem]>; // Scalar Unsigned Fixed-point Convert To Floating-Point (Immediate) -def int_aarch64_neon_vcvtf32_n_u32 : - Intrinsic<[llvm_float_ty], [llvm_v1i32_ty, llvm_i32_ty], [IntrNoMem]>; -def int_aarch64_neon_vcvtf64_n_u64 : - Intrinsic<[llvm_double_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>; +def int_aarch64_neon_vcvtfxu2fp_n : + Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty, llvm_i32_ty], [IntrNoMem]>; // Scalar Floating-point Convert To Signed Fixed-point (Immediate) -def int_aarch64_neon_vcvts_n_s32_f32 : - Intrinsic<[llvm_v1i32_ty], [llvm_v1f32_ty, llvm_i32_ty], [IntrNoMem]>; -def int_aarch64_neon_vcvtd_n_s64_f64 : - Intrinsic<[llvm_v1i64_ty], [llvm_v1f64_ty, llvm_i32_ty], [IntrNoMem]>; +def int_aarch64_neon_vcvtfp2fxs_n : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty, llvm_i32_ty], [IntrNoMem]>; // Scalar Floating-point Convert To Unsigned Fixed-point (Immediate) -def int_aarch64_neon_vcvts_n_u32_f32 : - Intrinsic<[llvm_v1i32_ty], [llvm_v1f32_ty, llvm_i32_ty], [IntrNoMem]>; -def int_aarch64_neon_vcvtd_n_u64_f64 : - Intrinsic<[llvm_v1i64_ty], [llvm_v1f64_ty, llvm_i32_ty], [IntrNoMem]>; - -class Neon_SHA_Intrinsic - : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v1i32_ty, llvm_v4i32_ty], - [IntrNoMem]>; +def int_aarch64_neon_vcvtfp2fxu_n : + Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty, llvm_i32_ty], [IntrNoMem]>; -def int_aarch64_neon_sha1c : Neon_SHA_Intrinsic; -def int_aarch64_neon_sha1m : Neon_SHA_Intrinsic; -def int_aarch64_neon_sha1p : Neon_SHA_Intrinsic; } |