https://gcc.gnu.org/g:7a9c30d132deee8a471c33100ed1924c8e5a6af3
commit r17-6163-g7a9c30d132deee8a471c33100ed1924c8e5a6af3 Author: Abhishek Kaushik <[email protected]> Date: Fri Nov 6 11:21:35 2026 +0000 aarch64: Port Neon table lookup intrinsics to new framework This patch ports the Neon table lookup intrinsics to the new pragma-based framework. Bootstrapped and regression tested on aarch64-linux-gnu. gcc/ChangeLog: * config/aarch64/aarch64-acle-builtins.cc (neon_function_groups): Handle DEF_NEON_FUNCTION_GS. * config/aarch64/aarch64-acle-builtins.h (TYPES_neon_tbl_s8): New. (TYPES_neon_tbl_other): Likewise. (TYPES_neon_luti2_b): Likewise. (TYPES_neon_luti2_h): Likewise. (DEF_SVE_TYPES_ARRAY): Add arrays for the new type sets. * config/aarch64/aarch64-neon-builtins-base.cc: Include emit-rtl.h. (gimple_table_lookup): New class. (gimple_qtbl): Likewise. (gimple_luti): Likewise. (vtbl1): Add function entries. (vtbl2): Likewise. (vtbl3): Likewise. (vtbl4): Likewise. (vqtbl1): Likewise. (vqtbl2): Likewise. (vqtbl3): Likewise. (vqtbl4): Likewise. (vqtbl1q): Likewise. (vqtbl2q): Likewise. (vqtbl3q): Likewise. (vqtbl4q): Likewise. (vtbx1): Likewise. (vtbx2): Likewise. (vtbx3): Likewise. (vtbx4): Likewise. (vqtbx1): Likewise. (vqtbx2): Likewise. (vqtbx3): Likewise. (vqtbx4): Likewise. (vqtbx1q): Likewise. (vqtbx2q): Likewise. (vqtbx3q): Likewise. (vqtbx4q): Likewise. (vluti2_lane): Likewise. (vluti2_laneq): Likewise. (vluti2q_lane): Likewise. (vluti2q_laneq): Likewise. (vluti4q_lane): Likewise. (vluti4q_laneq): Likewise. * config/aarch64/aarch64-neon-builtins-base.def: Add table lookup and LUTI entries. * config/aarch64/aarch64-neon-builtins.def (DEF_NEON_FUNCTION_GS): Define. * config/aarch64/aarch64-simd-pragma-builtins.def (ENTRY_TERNARY_VLUT8): Remove. (ENTRY_TERNARY_VLUT16): Likewise. Remove LUTI and table lookup entries. * config/aarch64/aarch64-simd.md (aarch64_qtbl<VB:mode><QTBL:mode>): New define_expand. (aarch64_qtbx<VB:mode><QTBL:mode>): Likewise. * config/aarch64/arm_neon.h: Remove hand-written table lookup and LUTI intrinsics. * config/aarch64/iterators.md (QTBL): New mode iterator. gcc/testsuite/ChangeLog: * gcc.target/aarch64/neon/vluti2.c: New test. * gcc.target/aarch64/neon/vluti4.c: New test. * gcc.target/aarch64/neon/vtbl.c: New test. * gcc.target/aarch64/neon/vtbx.c: New test. Diff: --- gcc/config/aarch64/aarch64-acle-builtins.cc | 8 +- gcc/config/aarch64/aarch64-acle-builtins.h | 18 + gcc/config/aarch64/aarch64-neon-builtins-base.cc | 203 +++++++ gcc/config/aarch64/aarch64-neon-builtins-base.def | 59 +++ gcc/config/aarch64/aarch64-neon-builtins.def | 7 + .../aarch64/aarch64-simd-pragma-builtins.def | 78 --- gcc/config/aarch64/aarch64-simd.md | 15 + gcc/config/aarch64/arm_neon.h | 588 --------------------- gcc/config/aarch64/iterators.md | 2 + gcc/testsuite/gcc.target/aarch64/neon/vluti2.c | 263 +++++++++ gcc/testsuite/gcc.target/aarch64/neon/vluti4.c | 138 +++++ gcc/testsuite/gcc.target/aarch64/neon/vtbl.c | 376 +++++++++++++ gcc/testsuite/gcc.target/aarch64/neon/vtbx.c | 405 ++++++++++++++ 13 files changed, 1492 insertions(+), 668 deletions(-) diff --git a/gcc/config/aarch64/aarch64-acle-builtins.cc b/gcc/config/aarch64/aarch64-acle-builtins.cc index ccfaeaafee49..bc2f24745d9c 100644 --- a/gcc/config/aarch64/aarch64-acle-builtins.cc +++ b/gcc/config/aarch64/aarch64-acle-builtins.cc @@ -188,17 +188,21 @@ constexpr group_suffix_info group_suffixes[] = { }; constexpr const aarch64_acle::function_group_info neon_function_groups[] = { -#define DEF_NEON_FUNCTION(NAME, TYPES, SHAPE_ARGS) \ +#define DEF_NEON_FUNCTION_GS(NAME, TYPES, GROUPS, SHAPE_ARGS) \ { \ /* .base_name = */ #NAME, \ /* .base = */ &aarch64_acle::functions::NAME, \ /* .shape = */ &aarch64_acle::shapes::SHAPE_NAME (NAME, TYPES),\ /* .types = */ aarch64_acle::types_##TYPES, \ - /* .groups = */ aarch64_acle::groups_none, \ + /* .groups = */ aarch64_acle::groups_##GROUPS, \ /* .preds = */ aarch64_acle::preds_none, \ /* .extensions = */ aarch64_required_extensions::REQUIRED_EXTENSIONS,\ /* .fpm_mode = */ aarch64_acle::FPM_unused, \ }, + +#define DEF_NEON_FUNCTION(NAME, TYPES, SHAPE_ARGS) \ + DEF_NEON_FUNCTION_GS (NAME, TYPES, none, SHAPE_ARGS) + #include "aarch64-neon-builtins.def" }; diff --git a/gcc/config/aarch64/aarch64-acle-builtins.h b/gcc/config/aarch64/aarch64-acle-builtins.h index 3260140f1e8e..bd07225f6c1d 100644 --- a/gcc/config/aarch64/aarch64-acle-builtins.h +++ b/gcc/config/aarch64/aarch64-acle-builtins.h @@ -2064,6 +2064,20 @@ function_expander::result_mode () const TYPES_neon_reinterpretq1 (D, p64), \ TYPES_neon_reinterpretq1 (D, p128) +/* _s8. */ +#define TYPES_neon_tbl_s8(S, D, T) \ + S (s8) + +/* _u8, _p8 and _mf8. */ +#define TYPES_neon_tbl_other(S, D, T) \ + S (u8), S (p8), S (mf8) + +#define TYPES_neon_luti2_b(S, D, T) \ + S (u8), S (s8), S (p8), S (mf8) + +#define TYPES_neon_luti2_h(S, D, T) \ + S (u16), S (s16), S (f16), S (bf16), S (p16) + /* Describe a tuple of type suffixes in which only the first is used. */ #define DEF_VECTOR_TYPE(X) \ { TYPE_SUFFIX_ ## X, NUM_TYPE_SUFFIXES, NUM_TYPE_SUFFIXES } @@ -2238,6 +2252,10 @@ DEF_SVE_TYPES_ARRAY (neon_rev64); DEF_SVE_TYPES_ARRAY (neon_reinterpret); DEF_SVE_TYPES_ARRAY (neon_reinterpretq); DEF_SVE_TYPES_ARRAY (neon_mul); +DEF_SVE_TYPES_ARRAY (neon_tbl_s8); +DEF_SVE_TYPES_ARRAY (neon_tbl_other); +DEF_SVE_TYPES_ARRAY (neon_luti2_b); +DEF_SVE_TYPES_ARRAY (neon_luti2_h); static const group_suffix_index groups_none[] = { GROUP_none, NUM_GROUP_SUFFIXES diff --git a/gcc/config/aarch64/aarch64-neon-builtins-base.cc b/gcc/config/aarch64/aarch64-neon-builtins-base.cc index 4dd26fd102da..4b66a93d0f01 100644 --- a/gcc/config/aarch64/aarch64-neon-builtins-base.cc +++ b/gcc/config/aarch64/aarch64-neon-builtins-base.cc @@ -44,6 +44,7 @@ #include "aarch64-neon-builtins-functions.h" #include "gimple-fold.h" #include "ssa.h" +#include "emit-rtl.h" namespace aarch64_acle { /* Build an asm volatile statement that prevents VALUE from being optimized @@ -749,6 +750,189 @@ public: } }; +/* Expand legacy NEON vtbl1-4 and vtbx1-4. + N is the number of 64-bit vectors in the lookup table. */ +template <unsigned int N> +class gimple_table_lookup : public gimple_function_base +{ + static_assert (N >= 1 && N <= 4, "Invalid number of table vectors"); + bool m_is_tbx; + +public: + constexpr gimple_table_lookup (bool tbx = false) + : m_is_tbx (tbx) + {} + + gimple *fold (gimple_folder &) const override { return nullptr; } + + rtx expand (function_expander &e) const override + { + insn_code m_code = (m_is_tbx && (N == 2 || N == 4) + ? (N == 2 ? CODE_FOR_aarch64_qtbx1v8qi + : CODE_FOR_aarch64_qtbx2v8qi) + : (N <= 2 ? CODE_FOR_aarch64_qtbl1v8qi + : CODE_FOR_aarch64_qtbl2v8qi)); + unsigned m_table_arg = m_is_tbx ? 1 : 0; + rtx parts[N]; + + /* vtbl1 and vtbx1 take a single V8QI vector. The remaining + intrinsics take a tuple of N V8QI vectors. */ + if (N == 1) + parts[0] = e.args[m_table_arg]; + else + { + machine_mode mode = GET_MODE (e.args[m_table_arg]); + + gcc_assert ((N == 2 && mode == V2x8QImode) + || (N == 3 && mode == V3x8QImode) + || (N == 4 && mode == V4x8QImode)); + + rtx tuple = force_reg (mode, e.args[m_table_arg]); + + /* Extract the individual 64-bit table vectors. */ + for (unsigned int i = 0; i < N; ++i) + parts[i] = simplify_gen_subreg (V8QImode, tuple, + mode, i * 8); + } + + /* Construct the full hardware table. N == 1 or 2 requires + one 128-bit register; N == 3 or 4 requires two. */ + rtx table = N > 2 ? gen_reg_rtx (V2x16QImode) + : gen_reg_rtx (V16QImode); + + /* Construct the first 128-bit table register. + For N == 1, zero the unused upper eight bytes. */ + rtx first = N > 2 ? simplify_gen_subreg (V16QImode, table, + V2x16QImode, 0) + : table; + + rtx upper = N >= 2 ? parts[1] + : CONST0_RTX (V8QImode); + + emit_insn (gen_aarch64_combinev8qi (first, + parts[0], upper)); + + if (N > 2) + { + rtx second = simplify_gen_subreg (V16QImode, table, + V2x16QImode, 16); + + upper = N == 4 ? parts[3] + : CONST0_RTX (V8QImode); + + emit_insn (gen_aarch64_combinev8qi (second, + parts[2], upper)); + } + + /* Legacy vtbx1 and vtbx3 have tables of 8 and 24 bytes, + respectively. Hardware TBX accepts tables whose sizes + are multiples of 16 bytes, so the padded bytes must + not overwrite the original destination. */ + if (m_is_tbx && (N == 1 || N == 3)) + { + rtx indices = force_reg (V8QImode, e.args[2]); + + /* Perform the table lookup. */ + rtx lookup = gen_reg_rtx (V8QImode); + + if (N == 1) + emit_insn (gen_aarch64_qtbl1v8qi (lookup, table, + indices)); + else + emit_insn (gen_aarch64_qtbl2v8qi (lookup, table, + indices)); + + /* Generate a mask for indices within the logical table. + The maximum valid index is 7 for vtbx1 and 23 for vtbx3. */ + rtx limit = gen_const_vec_duplicate (V8QImode, + GEN_INT (8 * N - 1)); + limit = force_reg (V8QImode, limit); + + rtx mask = gen_reg_rtx (V8QImode); + + /* mask[i] = 0xff if indices[i] <= 8 * N - 1. */ + emit_insn (gen_aarch64_cmgeuv8qi (mask, limit, indices)); + + /* Select lookup for valid indices, otherwise preserve the + original destination. */ + rtx result = gen_reg_rtx (V8QImode); + + rtx fallback = force_reg (V8QImode, e.args[0]); + + emit_insn (gen_aarch64_simd_bslv8qi (result, mask, + lookup, fallback)); + return result; + } + + /* All vtbl variants, and vtbx2/vtbx4, map directly onto + the corresponding TBL or TBX instruction after packing. */ + e.args[m_table_arg] = table; + return e.use_exact_insn (m_code); + } +}; + +template <bool TBX> +class gimple_qtbl : public gimple_function_base +{ +public: + gimple *fold (gimple_folder &) const override { return nullptr; } + + rtx expand (function_expander &e) const override + { + constexpr unsigned int table_arg = TBX ? 1 : 0; + insn_code icode + = TBX ? code_for_aarch64_qtbx (e.result_mode (), + GET_MODE (e.args[table_arg])) + : code_for_aarch64_qtbl (e.result_mode (), + GET_MODE (e.args[table_arg])); + + return e.use_exact_insn (icode); + } +}; + +/* Expand NEON LUTI2 and LUTI4 intrinsics. + BITS_PER_INDEX is the number of bits per lookup index. */ +template <unsigned int BITS_PER_INDEX> +class gimple_luti : public gimple_function_base +{ + static_assert (BITS_PER_INDEX == 2 || BITS_PER_INDEX == 4); + +public: + gimple *fold (gimple_folder &) const override { return nullptr; } + + rtx expand (function_expander &e) const override + { + rtx table = e.args[0]; + + /* LUTI2 uses only the low 64 bits of the table. + The 128-bit table forms are provided for convenience. */ + if (BITS_PER_INDEX == 2 + && known_eq (GET_MODE_BITSIZE (GET_MODE (table)), 128u)) + { + machine_mode mode + = aarch64_v64_mode (GET_MODE_INNER (GET_MODE (table))) + .require (); + + table = gen_lowpart (mode, table); + } + + /* Select the instruction based on the table and index modes. */ + insn_code icode + = code_for_aarch64_lut (GET_MODE (table), + GET_MODE (e.args[1])); + + gcc_assert (icode != CODE_FOR_nothing); + + e.add_output_operand (icode); + e.add_input_operand (icode, table); + e.add_input_operand (icode, e.args[1]); + e.add_input_operand (icode, e.args[2]); + e.add_integer_operand (BITS_PER_INDEX); + + return e.generate_insn (icode); + } +}; + tree ext_mask (gimple_folder &f) { @@ -1064,6 +1248,25 @@ NEON_FUNCTION (vzip1, vzip1q, gimple_permute, (zip_mask<false>)) NEON_FUNCTION (vzip2, vzip2q, gimple_permute, (zip_mask<true>)) NEON_FUNCTION (vzip, vzipq, gimple_permute_pair, (zip_mask<false>, zip_mask<true>)) +// Table lookup +NEON_FUNCTION (vtbl1, gimple_table_lookup<1>,) +NEON_FUNCTION (vtbl2, gimple_table_lookup<2>,) +NEON_FUNCTION (vtbl3, gimple_table_lookup<3>,) +NEON_FUNCTION (vtbl4, gimple_table_lookup<4>,) +NEON_FUNCTION (vqtbl1, vqtbl2, vqtbl3, vqtbl4, + vqtbl1q, vqtbl2q, vqtbl3q, vqtbl4q, gimple_qtbl<false>,) +NEON_FUNCTION (vtbx1, gimple_table_lookup<1>, (true)) +NEON_FUNCTION (vtbx2, gimple_table_lookup<2>, (true)) +NEON_FUNCTION (vtbx3, gimple_table_lookup<3>, (true)) +NEON_FUNCTION (vtbx4, gimple_table_lookup<4>, (true)) +NEON_FUNCTION (vqtbx1, vqtbx2, vqtbx3, vqtbx4, + vqtbx1q, vqtbx2q, vqtbx3q, vqtbx4q, gimple_qtbl<true>,) + +// Table lookup with 2-bit or 4-bit indices +NEON_FUNCTION (vluti2_lane, vluti2_laneq, + vluti2q_lane, vluti2q_laneq, gimple_luti<2>,) +NEON_FUNCTION (vluti4q_lane, vluti4q_laneq, gimple_luti<4>,) + // AES NEON_FUNCTION (vaeseq, gimple_exact_insn<CODE_FOR_aarch64_crypto_aesev16qi>,) NEON_FUNCTION (vaesdq, gimple_exact_insn<CODE_FOR_aarch64_crypto_aesdv16qi>,) diff --git a/gcc/config/aarch64/aarch64-neon-builtins-base.def b/gcc/config/aarch64/aarch64-neon-builtins-base.def index a42a25135a0c..db9defc454c9 100644 --- a/gcc/config/aarch64/aarch64-neon-builtins-base.def +++ b/gcc/config/aarch64/aarch64-neon-builtins-base.def @@ -346,6 +346,65 @@ DEF_NEON_FUNCTION (vzip2q, bhsd_neon, ("Q0,Q0,Q0")) DEF_NEON_FUNCTION (vzipq, bhs_neon, ("Q0x2,Q0,Q0")) #undef REQUIRED_EXTENSIONS +// Table lookup +#define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_SIMD) +DEF_NEON_FUNCTION (vtbl1, neon_tbl_s8, ("D0,D0,D0")) +DEF_NEON_FUNCTION (vtbl1, neon_tbl_other, ("D0,D0,Du0")) +DEF_NEON_FUNCTION (vtbl2, neon_tbl_s8, ("D0,D0x2,D0")) +DEF_NEON_FUNCTION (vtbl2, neon_tbl_other, ("D0,D0x2,Du0")) +DEF_NEON_FUNCTION (vtbl3, neon_tbl_s8, ("D0,D0x3,D0")) +DEF_NEON_FUNCTION (vtbl3, neon_tbl_other, ("D0,D0x3,Du0")) +DEF_NEON_FUNCTION (vtbl4, neon_tbl_s8, ("D0,D0x4,D0")) +DEF_NEON_FUNCTION (vtbl4, neon_tbl_other, ("D0,D0x4,Du0")) +DEF_NEON_FUNCTION (vqtbl1, neon_rev16, ("D0,Q0,Du0")) +DEF_NEON_FUNCTION (vqtbl2, neon_rev16, ("D0,Q0x2,Du0")) +DEF_NEON_FUNCTION (vqtbl3, neon_rev16, ("D0,Q0x3,Du0")) +DEF_NEON_FUNCTION (vqtbl4, neon_rev16, ("D0,Q0x4,Du0")) +DEF_NEON_FUNCTION (vqtbl1q, neon_rev16, ("Q0,Q0,Qu0")) +DEF_NEON_FUNCTION (vqtbl2q, neon_rev16, ("Q0,Q0x2,Qu0")) +DEF_NEON_FUNCTION (vqtbl3q, neon_rev16, ("Q0,Q0x3,Qu0")) +DEF_NEON_FUNCTION (vqtbl4q, neon_rev16, ("Q0,Q0x4,Qu0")) +DEF_NEON_FUNCTION (vtbx1, neon_tbl_s8, ("D0,D0,D0,D0")) +DEF_NEON_FUNCTION (vtbx1, neon_tbl_other, ("D0,D0,D0,Du0")) +DEF_NEON_FUNCTION (vtbx2, neon_tbl_s8, ("D0,D0,D0x2,D0")) +DEF_NEON_FUNCTION (vtbx2, neon_tbl_other, ("D0,D0,D0x2,Du0")) +DEF_NEON_FUNCTION (vtbx3, neon_tbl_s8, ("D0,D0,D0x3,D0")) +DEF_NEON_FUNCTION (vtbx3, neon_tbl_other, ("D0,D0,D0x3,Du0")) +DEF_NEON_FUNCTION (vtbx4, neon_tbl_s8, ("D0,D0,D0x4,D0")) +DEF_NEON_FUNCTION (vtbx4, neon_tbl_other, ("D0,D0,D0x4,Du0")) +DEF_NEON_FUNCTION (vqtbx1, neon_rev16, ("D0,D0,Q0,Du0")) +DEF_NEON_FUNCTION (vqtbx2, neon_rev16, ("D0,D0,Q0x2,Du0")) +DEF_NEON_FUNCTION (vqtbx3, neon_rev16, ("D0,D0,Q0x3,Du0")) +DEF_NEON_FUNCTION (vqtbx4, neon_rev16, ("D0,D0,Q0x4,Du0")) +DEF_NEON_FUNCTION (vqtbx1q, neon_rev16, ("Q0,Q0,Q0,Qu0")) +DEF_NEON_FUNCTION (vqtbx2q, neon_rev16, ("Q0,Q0,Q0x2,Qu0")) +DEF_NEON_FUNCTION (vqtbx3q, neon_rev16, ("Q0,Q0,Q0x3,Qu0")) +DEF_NEON_FUNCTION (vqtbx4q, neon_rev16, ("Q0,Q0,Q0x4,Qu0")) +#undef REQUIRED_EXTENSIONS + +/* Lookup table read with packed 2-bit indices. */ +#define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_LUT) +/* 8-bit result elements. */ +DEF_NEON_FUNCTION (vluti2_lane, neon_luti2_b, ("Q0,D0,Du8,ss32", range<2, 0, 1>)) +DEF_NEON_FUNCTION (vluti2_laneq, neon_luti2_b, ("Q0,D0,Qu8,ss32", range<2, 0, 3>)) +DEF_NEON_FUNCTION (vluti2q_lane, neon_luti2_b, ("Q0,Q0,Du8,ss32", range<2, 0, 1>)) +DEF_NEON_FUNCTION (vluti2q_laneq, neon_luti2_b, ("Q0,Q0,Qu8,ss32", range<2, 0, 3>)) + +/* 16-bit result elements. */ +DEF_NEON_FUNCTION (vluti2_lane, neon_luti2_h, ("Q0,D0,Du8,ss32", range<2, 0, 3>)) +DEF_NEON_FUNCTION (vluti2_laneq, neon_luti2_h, ("Q0,D0,Qu8,ss32", range<2, 0, 7>)) +DEF_NEON_FUNCTION (vluti2q_lane, neon_luti2_h, ("Q0,Q0,Du8,ss32", range<2, 0, 3>)) +DEF_NEON_FUNCTION (vluti2q_laneq, neon_luti2_h, ("Q0,Q0,Qu8,ss32", range<2, 0, 7>)) + +/* 8-bit forms: no group suffix. */ +DEF_NEON_FUNCTION (vluti4q_lane, neon_luti2_b, ("Q0,Q0,Du8,ss32", range<2, 0, 0>)) +DEF_NEON_FUNCTION (vluti4q_laneq, neon_luti2_b, ("Q0,Q0,Qu8,ss32", range<2, 0, 1>)) + +/* 16-bit forms: the name ends with _x2. */ +DEF_NEON_FUNCTION_GS (vluti4q_lane, neon_luti2_h, x2, ("Q0,Q0x2,Du8,ss32", range<2, 0, 1>)) +DEF_NEON_FUNCTION_GS (vluti4q_laneq, neon_luti2_h, x2, ("Q0,Q0x2,Qu8,ss32", range<2, 0, 3>)) +#undef REQUIRED_EXTENSIONS + // AES #define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_SIMD | AARCH64_FL_AES) DEF_NEON_FUNCTION (vaeseq, b_unsigned, ("Q0,Q0,Q0")) diff --git a/gcc/config/aarch64/aarch64-neon-builtins.def b/gcc/config/aarch64/aarch64-neon-builtins.def index 58630eebe489..02bb5591b8ce 100644 --- a/gcc/config/aarch64/aarch64-neon-builtins.def +++ b/gcc/config/aarch64/aarch64-neon-builtins.def @@ -25,6 +25,12 @@ #define DEF_NEON_FUNCTION(NAME, TYPES, SHAPE_ARGS) #endif +/* Define a NEON function with an explicit group suffix. */ +#ifndef DEF_NEON_FUNCTION_GS +#define DEF_NEON_FUNCTION_GS(NAME, TYPES, GROUPS, SHAPE_ARGS) \ + DEF_NEON_FUNCTION (NAME, TYPES, SHAPE_ARGS) +#endif + /* Helper for generating the name of the function_group's corresponding neon_shape instance. */ #define OBJ_NAME(NAME, TYPES) NAME ## _ ## TYPES ## _obj @@ -36,5 +42,6 @@ #include "aarch64-neon-builtins-base.def" #undef DEF_NEON_FUNCTION +#undef DEF_NEON_FUNCTION_GS #undef OBJ_NAME #undef SHAPE_NAME diff --git a/gcc/config/aarch64/aarch64-simd-pragma-builtins.def b/gcc/config/aarch64/aarch64-simd-pragma-builtins.def index 92ae493800ea..6bd4e8e89714 100644 --- a/gcc/config/aarch64/aarch64-simd-pragma-builtins.def +++ b/gcc/config/aarch64/aarch64-simd-pragma-builtins.def @@ -78,36 +78,6 @@ ENTRY_BINARY (NAME##q_f32, f32q, f32q, s32q, UNSPEC, FLAGS) \ ENTRY_BINARY (NAME##q_f64, f64q, f64q, s64q, UNSPEC, FLAGS) -#undef ENTRY_TERNARY_VLUT8 -#define ENTRY_TERNARY_VLUT8(T) \ - ENTRY_BINARY_LANE (vluti2_lane_##T##8, T##8q, T##8, u8, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti2_laneq_##T##8, T##8q, T##8, u8q, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti2q_lane_##T##8, T##8q, T##8q, u8, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti2q_laneq_##T##8, T##8q, T##8q, u8q, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti4q_lane_##T##8, T##8q, T##8q, u8, \ - UNSPEC_LUTI4, QUIET) \ - ENTRY_BINARY_LANE (vluti4q_laneq_##T##8, T##8q, T##8q, u8q, \ - UNSPEC_LUTI4, QUIET) - -#undef ENTRY_TERNARY_VLUT16 -#define ENTRY_TERNARY_VLUT16(T) \ - ENTRY_BINARY_LANE (vluti2_lane_##T##16, T##16q, T##16, u8, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti2_laneq_##T##16, T##16q, T##16, u8q, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti2q_lane_##T##16, T##16q, T##16q, u8, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti2q_laneq_##T##16, T##16q, T##16q, u8q, \ - UNSPEC_LUTI2, QUIET) \ - ENTRY_BINARY_LANE (vluti4q_lane_##T##16_x2, T##16q, T##16qx2, u8, \ - UNSPEC_LUTI4, QUIET) \ - ENTRY_BINARY_LANE (vluti4q_laneq_##T##16_x2, T##16q, T##16qx2, u8q, \ - UNSPEC_LUTI4, QUIET) - #undef ENTRY_UNARY_VQ_BHF #define ENTRY_UNARY_VQ_BHF(N, T1, UNSPEC, FLAGS) \ ENTRY_UNARY (N##_bf16_mf8_fpm, bf16q, T1, UNSPEC, FLAGS) \ @@ -140,20 +110,6 @@ ENTRY_BINARY_VHSDF (vamax, UNSPEC_FAMAX, FP) ENTRY_BINARY_VHSDF (vamin, UNSPEC_FAMIN, FP) #undef REQUIRED_EXTENSIONS -// lut -#define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_LUT) -ENTRY_TERNARY_VLUT8 (p) -ENTRY_TERNARY_VLUT8 (s) -ENTRY_TERNARY_VLUT8 (u) -ENTRY_TERNARY_VLUT8 (mf) - -ENTRY_TERNARY_VLUT16 (bf) -ENTRY_TERNARY_VLUT16 (f) -ENTRY_TERNARY_VLUT16 (p) -ENTRY_TERNARY_VLUT16 (s) -ENTRY_TERNARY_VLUT16 (u) -#undef REQUIRED_EXTENSIONS - // fpm conversion #define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_FP8) ENTRY_UNARY_VQ_BHF (vcvt1, mf8, UNSPEC_F1CVTL_FP8, FP8) @@ -287,37 +243,3 @@ ENTRY_STORE (vst1q_mf8_x4, mf8_scalar_ptr, mf8qx4, UNSPEC_ST1x4) ENTRY_STORE_LANE (vst4_lane_mf8, mf8_scalar_ptr, mf8x4, UNSPEC_ST4_LANE) ENTRY_STORE_LANE (vst4q_lane_mf8, mf8_scalar_ptr, mf8qx4, UNSPEC_ST4_LANE) #undef REQUIRED_EXTENSIONS - -// tbl<n> -#define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_SIMD) -ENTRY_BINARY (vtbl1_mf8, mf8, mf8, u8, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vtbl2_mf8, mf8, mf8x2, u8, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vtbl3_mf8, mf8, mf8x3, u8, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vtbl4_mf8, mf8, mf8x4, u8, UNSPEC_TBL, QUIET) - -ENTRY_BINARY (vqtbl1_mf8, mf8, mf8q, u8, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vqtbl1q_mf8, mf8q, mf8q, u8q, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vqtbl2_mf8, mf8, mf8qx2, u8, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vqtbl2q_mf8, mf8q, mf8qx2, u8q, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vqtbl3_mf8, mf8, mf8qx3, u8, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vqtbl3q_mf8, mf8q, mf8qx3, u8q, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vqtbl4_mf8, mf8, mf8qx4, u8, UNSPEC_TBL, QUIET) -ENTRY_BINARY (vqtbl4q_mf8, mf8q, mf8qx4, u8q, UNSPEC_TBL, QUIET) -#undef REQUIRED_EXTENSIONS - -// tbx<n> -#define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_SIMD) -ENTRY_TERNARY (vtbx1_mf8, mf8, mf8, mf8, u8, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vtbx2_mf8, mf8, mf8, mf8x2, u8, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vtbx3_mf8, mf8, mf8, mf8x3, u8, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vtbx4_mf8, mf8, mf8, mf8x4, u8, UNSPEC_TBX, QUIET) - -ENTRY_TERNARY (vqtbx1_mf8, mf8, mf8, mf8q, u8, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vqtbx1q_mf8, mf8q, mf8q, mf8q, u8q, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vqtbx2_mf8, mf8, mf8, mf8qx2, u8, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vqtbx2q_mf8, mf8q, mf8q, mf8qx2, u8q, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vqtbx3_mf8, mf8, mf8, mf8qx3, u8, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vqtbx3q_mf8, mf8q, mf8q, mf8qx3, u8q, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vqtbx4_mf8, mf8, mf8, mf8qx4, u8, UNSPEC_TBX, QUIET) -ENTRY_TERNARY (vqtbx4q_mf8, mf8q, mf8q, mf8qx4, u8q, UNSPEC_TBX, QUIET) -#undef REQUIRED_EXTENSIONS diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md index 11705bc90738..c589ee21b89f 100644 --- a/gcc/config/aarch64/aarch64-simd.md +++ b/gcc/config/aarch64/aarch64-simd.md @@ -9475,6 +9475,21 @@ DONE; }) +(define_expand "@aarch64_qtbl<VB:mode><QTBL:mode>" + [(set (match_operand:VB 0 "register_operand") + (unspec:VB [(match_operand:QTBL 1 "register_operand") + (match_operand:VB 2 "register_operand")] + UNSPEC_TBL))] + "TARGET_SIMD") + +(define_expand "@aarch64_qtbx<VB:mode><QTBL:mode>" + [(set (match_operand:VB 0 "register_operand") + (unspec:VB [(match_operand:VB 1 "register_operand") + (match_operand:QTBL 2 "register_operand") + (match_operand:VB 3 "register_operand")] + UNSPEC_TBX))] + "TARGET_SIMD") + (define_insn "aarch64_qtbl1<mode>" [(set (match_operand:VB 0 "register_operand" "=w") (unspec:VB [(match_operand:V16QI 1 "register_operand" "w") diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h index 9caa0a2088d8..a0dca5e8da9e 100644 --- a/gcc/config/aarch64/arm_neon.h +++ b/gcc/config/aarch64/arm_neon.h @@ -4066,232 +4066,6 @@ vqrdmulhq_laneq_s32 (int32x4_t __a, int32x4_t __b, const int __c) return __builtin_aarch64_sqrdmulh_laneqv4si (__a, __b, __c); } -/* Table intrinsics. */ - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl1_p8 (poly8x16_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl1v8qi_ppu (__tab, __idx); -} - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl1_s8 (int8x16_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl1v8qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl1_u8 (uint8x16_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl1v8qi_uuu (__tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl1q_p8 (poly8x16_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl1v16qi_ppu (__tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl1q_s8 (int8x16_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl1v16qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl1q_u8 (uint8x16_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl1v16qi_uuu (__tab, __idx); -} - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx1_s8 (int8x8_t __r, int8x16_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx1v8qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx1_u8 (uint8x8_t __r, uint8x16_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx1v8qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx1_p8 (poly8x8_t __r, poly8x16_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx1v8qi_pppu (__r, __tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx1q_s8 (int8x16_t __r, int8x16_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx1v16qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx1q_u8 (uint8x16_t __r, uint8x16_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx1v16qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx1q_p8 (poly8x16_t __r, poly8x16_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx1v16qi_pppu (__r, __tab, __idx); -} - -/* V7 legacy table intrinsics. */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl1_s8 (int8x8_t __tab, int8x8_t __idx) -{ - int8x16_t __temp = vcombine_s8 (__tab, - vcreate_s8 (__AARCH64_UINT64_C (0x0))); - return __builtin_aarch64_qtbl1v8qi (__temp, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl1_u8 (uint8x8_t __tab, uint8x8_t __idx) -{ - uint8x16_t __temp = vcombine_u8 (__tab, - vcreate_u8 (__AARCH64_UINT64_C (0x0))); - return __builtin_aarch64_qtbl1v8qi_uuu (__temp, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl1_p8 (poly8x8_t __tab, uint8x8_t __idx) -{ - poly8x16_t __temp = vcombine_p8 (__tab, - vcreate_p8 (__AARCH64_UINT64_C (0x0))); - return __builtin_aarch64_qtbl1v8qi_ppu (__temp, __idx); -} - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl2_s8 (int8x8x2_t __tab, int8x8_t __idx) -{ - int8x16_t __temp = vcombine_s8 (__tab.val[0], __tab.val[1]); - return __builtin_aarch64_qtbl1v8qi (__temp, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl2_u8 (uint8x8x2_t __tab, uint8x8_t __idx) -{ - uint8x16_t __temp = vcombine_u8 (__tab.val[0], __tab.val[1]); - return __builtin_aarch64_qtbl1v8qi_uuu (__temp, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl2_p8 (poly8x8x2_t __tab, uint8x8_t __idx) -{ - poly8x16_t __temp = vcombine_p8 (__tab.val[0], __tab.val[1]); - return __builtin_aarch64_qtbl1v8qi_ppu (__temp, __idx); -} - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl3_s8 (int8x8x3_t __tab, int8x8_t __idx) -{ - int8x16x2_t __temp; - __temp.val[0] = vcombine_s8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_s8 (__tab.val[2], - vcreate_s8 (__AARCH64_UINT64_C (0x0))); - return __builtin_aarch64_qtbl2v8qi (__temp, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl3_u8 (uint8x8x3_t __tab, uint8x8_t __idx) -{ - uint8x16x2_t __temp; - __temp.val[0] = vcombine_u8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_u8 (__tab.val[2], - vcreate_u8 (__AARCH64_UINT64_C (0x0))); - return __builtin_aarch64_qtbl2v8qi_uuu (__temp, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl3_p8 (poly8x8x3_t __tab, uint8x8_t __idx) -{ - poly8x16x2_t __temp; - __temp.val[0] = vcombine_p8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_p8 (__tab.val[2], - vcreate_p8 (__AARCH64_UINT64_C (0x0))); - return __builtin_aarch64_qtbl2v8qi_ppu (__temp, __idx); -} - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl4_s8 (int8x8x4_t __tab, int8x8_t __idx) -{ - int8x16x2_t __temp; - __temp.val[0] = vcombine_s8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_s8 (__tab.val[2], __tab.val[3]); - return __builtin_aarch64_qtbl2v8qi (__temp, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl4_u8 (uint8x8x4_t __tab, uint8x8_t __idx) -{ - uint8x16x2_t __temp; - __temp.val[0] = vcombine_u8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_u8 (__tab.val[2], __tab.val[3]); - return __builtin_aarch64_qtbl2v8qi_uuu (__temp, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbl4_p8 (poly8x8x4_t __tab, uint8x8_t __idx) -{ - poly8x16x2_t __temp; - __temp.val[0] = vcombine_p8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_p8 (__tab.val[2], __tab.val[3]); - return __builtin_aarch64_qtbl2v8qi_ppu (__temp, __idx); -} - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx2_s8 (int8x8_t __r, int8x8x2_t __tab, int8x8_t __idx) -{ - int8x16_t __temp = vcombine_s8 (__tab.val[0], __tab.val[1]); - return __builtin_aarch64_qtbx1v8qi (__r, __temp, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx2_u8 (uint8x8_t __r, uint8x8x2_t __tab, uint8x8_t __idx) -{ - uint8x16_t __temp = vcombine_u8 (__tab.val[0], __tab.val[1]); - return __builtin_aarch64_qtbx1v8qi_uuuu (__r, __temp, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx2_p8 (poly8x8_t __r, poly8x8x2_t __tab, uint8x8_t __idx) -{ - poly8x16_t __temp = vcombine_p8 (__tab.val[0], __tab.val[1]); - return __builtin_aarch64_qtbx1v8qi_pppu (__r, __temp, __idx); -} - /* End of temporary inline asm. */ /* Start of optimal implementations in approved order. */ @@ -13571,270 +13345,6 @@ vqshrund_n_s64 (int64_t __a, const int __b) return (int32_t) __builtin_aarch64_sqshrun_ndi (__a, __b); } -/* vqtbl2 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl2_s8 (int8x16x2_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl2v8qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl2_u8 (uint8x16x2_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl2v8qi_uuu (__tab, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl2_p8 (poly8x16x2_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl2v8qi_ppu (__tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl2q_s8 (int8x16x2_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl2v16qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl2q_u8 (uint8x16x2_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl2v16qi_uuu (__tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl2q_p8 (poly8x16x2_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl2v16qi_ppu (__tab, __idx); -} - -/* vqtbl3 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl3_s8 (int8x16x3_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl3v8qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl3_u8 (uint8x16x3_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl3v8qi_uuu (__tab, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl3_p8 (poly8x16x3_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl3v8qi_ppu (__tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl3q_s8 (int8x16x3_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl3v16qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl3q_u8 (uint8x16x3_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl3v16qi_uuu (__tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl3q_p8 (poly8x16x3_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl3v16qi_ppu (__tab, __idx); -} - -/* vqtbl4 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl4_s8 (int8x16x4_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl4v8qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl4_u8 (uint8x16x4_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl4v8qi_uuu (__tab, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl4_p8 (poly8x16x4_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbl4v8qi_ppu (__tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl4q_s8 (int8x16x4_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl4v16qi_ssu (__tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl4q_u8 (uint8x16x4_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl4v16qi_uuu (__tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbl4q_p8 (poly8x16x4_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbl4v16qi_ppu (__tab, __idx); -} - -/* vqtbx2 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx2_s8 (int8x8_t __r, int8x16x2_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx2v8qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx2_u8 (uint8x8_t __r, uint8x16x2_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx2v8qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx2_p8 (poly8x8_t __r, poly8x16x2_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx2v8qi_pppu (__r, __tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx2q_s8 (int8x16_t __r, int8x16x2_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx2v16qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx2q_u8 (uint8x16_t __r, uint8x16x2_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx2v16qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx2q_p8 (poly8x16_t __r, poly8x16x2_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx2v16qi_pppu (__r, __tab, __idx); -} - -/* vqtbx3 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx3_s8 (int8x8_t __r, int8x16x3_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx3v8qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx3_u8 (uint8x8_t __r, uint8x16x3_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx3v8qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx3_p8 (poly8x8_t __r, poly8x16x3_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx3v8qi_pppu (__r, __tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx3q_s8 (int8x16_t __r, int8x16x3_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx3v16qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx3q_u8 (uint8x16_t __r, uint8x16x3_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx3v16qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx3q_p8 (poly8x16_t __r, poly8x16x3_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx3v16qi_pppu (__r, __tab, __idx); -} - -/* vqtbx4 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx4_s8 (int8x8_t __r, int8x16x4_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx4v8qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx4_u8 (uint8x8_t __r, uint8x16x4_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx4v8qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx4_p8 (poly8x8_t __r, poly8x16x4_t __tab, uint8x8_t __idx) -{ - return __builtin_aarch64_qtbx4v8qi_pppu (__r, __tab, __idx); -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx4q_s8 (int8x16_t __r, int8x16x4_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx4v16qi_sssu (__r, __tab, __idx); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx4q_u8 (uint8x16_t __r, uint8x16x4_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx4v16qi_uuuu (__r, __tab, __idx); -} - -__extension__ extern __inline poly8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vqtbx4q_p8 (poly8x16_t __r, poly8x16x4_t __tab, uint8x16_t __idx) -{ - return __builtin_aarch64_qtbx4v16qi_pppu (__r, __tab, __idx); -} - /* vrecpe */ __extension__ extern __inline uint32x2_t @@ -16985,104 +16495,6 @@ vsubd_u64 (uint64_t __a, uint64_t __b) return __a - __b; } -/* vtbx1 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx1_s8 (int8x8_t __r, int8x8_t __tab, int8x8_t __idx) -{ - uint8x8_t __mask = vclt_u8 (vreinterpret_u8_s8 (__idx), - vmov_n_u8 (8)); - int8x8_t __tbl = vtbl1_s8 (__tab, __idx); - - return vbsl_s8 (__mask, __tbl, __r); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx1_u8 (uint8x8_t __r, uint8x8_t __tab, uint8x8_t __idx) -{ - uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (8)); - uint8x8_t __tbl = vtbl1_u8 (__tab, __idx); - - return vbsl_u8 (__mask, __tbl, __r); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx1_p8 (poly8x8_t __r, poly8x8_t __tab, uint8x8_t __idx) -{ - uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (8)); - poly8x8_t __tbl = vtbl1_p8 (__tab, __idx); - - return vbsl_p8 (__mask, __tbl, __r); -} - -/* vtbx3 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx3_s8 (int8x8_t __r, int8x8x3_t __tab, int8x8_t __idx) -{ - uint8x8_t __mask = vclt_u8 (vreinterpret_u8_s8 (__idx), - vmov_n_u8 (24)); - int8x8_t __tbl = vtbl3_s8 (__tab, __idx); - - return vbsl_s8 (__mask, __tbl, __r); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx3_u8 (uint8x8_t __r, uint8x8x3_t __tab, uint8x8_t __idx) -{ - uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (24)); - uint8x8_t __tbl = vtbl3_u8 (__tab, __idx); - - return vbsl_u8 (__mask, __tbl, __r); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx3_p8 (poly8x8_t __r, poly8x8x3_t __tab, uint8x8_t __idx) -{ - uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (24)); - poly8x8_t __tbl = vtbl3_p8 (__tab, __idx); - - return vbsl_p8 (__mask, __tbl, __r); -} - -/* vtbx4 */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx4_s8 (int8x8_t __r, int8x8x4_t __tab, int8x8_t __idx) -{ - int8x16x2_t __temp; - __temp.val[0] = vcombine_s8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_s8 (__tab.val[2], __tab.val[3]); - return __builtin_aarch64_qtbx2v8qi (__r, __temp, __idx); -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx4_u8 (uint8x8_t __r, uint8x8x4_t __tab, uint8x8_t __idx) -{ - uint8x16x2_t __temp; - __temp.val[0] = vcombine_u8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_u8 (__tab.val[2], __tab.val[3]); - return __builtin_aarch64_qtbx2v8qi_uuuu (__r, __temp, __idx); -} - -__extension__ extern __inline poly8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vtbx4_p8 (poly8x8_t __r, poly8x8x4_t __tab, uint8x8_t __idx) -{ - poly8x16x2_t __temp; - __temp.val[0] = vcombine_p8 (__tab.val[0], __tab.val[1]); - __temp.val[1] = vcombine_p8 (__tab.val[2], __tab.val[3]); - return __builtin_aarch64_qtbx2v8qi_pppu (__r, __temp, __idx); -} - /* vtst */ __extension__ extern __inline uint8x8_t diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md index 30bbd5878b6d..f69bf88a47d4 100644 --- a/gcc/config/aarch64/iterators.md +++ b/gcc/config/aarch64/iterators.md @@ -476,6 +476,8 @@ ;; Advanced SIMD 128-bit vector structure modes. (define_mode_iterator VSTRUCT_Q [VSTRUCT_2Q VSTRUCT_3Q VSTRUCT_4Q]) +(define_mode_iterator QTBL [V16QI V2x16QI V3x16QI V4x16QI]) + ;; Advanced SIMD 2-vector structure modes. (define_mode_iterator VSTRUCT_2QD [VSTRUCT_2D VSTRUCT_2Q]) diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vluti2.c b/gcc/testsuite/gcc.target/aarch64/neon/vluti2.c new file mode 100644 index 000000000000..9a0fef0ab359 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vluti2.c @@ -0,0 +1,263 @@ +/* { dg-do compile } */ +/* { dg-additional-options "-O3 -march=armv9-a+lut" } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +#define TEST_LUTI2(NAME, RET_TYPE, ARG_1_TYPE, ARG_2_TYPE) \ + RET_TYPE test_##NAME (ARG_1_TYPE vn, ARG_2_TYPE vm) \ + { \ + return NAME (vn, vm, 0); \ + } + +/* +** test_vluti2_lane_u8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_u8, uint8x16_t, uint8x8_t, uint8x8_t) + +/* +** test_vluti2_laneq_u8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_u8, uint8x16_t, uint8x8_t, uint8x16_t) + +/* +** test_vluti2q_lane_u8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_u8, uint8x16_t, uint8x16_t, uint8x8_t) + +/* +** test_vluti2q_laneq_u8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_u8, uint8x16_t, uint8x16_t, uint8x16_t) + +/* +** test_vluti2_lane_s8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_s8, int8x16_t, int8x8_t, uint8x8_t) + +/* +** test_vluti2_laneq_s8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_s8, int8x16_t, int8x8_t, uint8x16_t) + +/* +** test_vluti2q_lane_s8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_s8, int8x16_t, int8x16_t, uint8x8_t) + +/* +** test_vluti2q_laneq_s8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_s8, int8x16_t, int8x16_t, uint8x16_t) + +/* +** test_vluti2_lane_p8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_p8, poly8x16_t, poly8x8_t, uint8x8_t) + +/* +** test_vluti2_laneq_p8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_p8, poly8x16_t, poly8x8_t, uint8x16_t) + +/* +** test_vluti2q_lane_p8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_p8, poly8x16_t, poly8x16_t, uint8x8_t) + +/* +** test_vluti2q_laneq_p8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_p8, poly8x16_t, poly8x16_t, uint8x16_t) + +/* +** test_vluti2_lane_mf8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_mf8, mfloat8x16_t, mfloat8x8_t, uint8x8_t) + +/* +** test_vluti2_laneq_mf8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_mf8, mfloat8x16_t, mfloat8x8_t, uint8x16_t) + +/* +** test_vluti2q_lane_mf8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_mf8, mfloat8x16_t, mfloat8x16_t, uint8x8_t) + +/* +** test_vluti2q_laneq_mf8: +** luti2 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_mf8, mfloat8x16_t, mfloat8x16_t, uint8x16_t) + +/* +** test_vluti2_lane_u16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_u16, uint16x8_t, uint16x4_t, uint8x8_t) + +/* +** test_vluti2_laneq_u16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_u16, uint16x8_t, uint16x4_t, uint8x16_t) + +/* +** test_vluti2q_lane_u16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_u16, uint16x8_t, uint16x8_t, uint8x8_t) + +/* +** test_vluti2q_laneq_u16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_u16, uint16x8_t, uint16x8_t, uint8x16_t) + +/* +** test_vluti2_lane_s16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_s16, int16x8_t, int16x4_t, uint8x8_t) + +/* +** test_vluti2_laneq_s16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_s16, int16x8_t, int16x4_t, uint8x16_t) + +/* +** test_vluti2q_lane_s16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_s16, int16x8_t, int16x8_t, uint8x8_t) + +/* +** test_vluti2q_laneq_s16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_s16, int16x8_t, int16x8_t, uint8x16_t) + +/* +** test_vluti2_lane_f16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_f16, float16x8_t, float16x4_t, uint8x8_t) + +/* +** test_vluti2_laneq_f16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_f16, float16x8_t, float16x4_t, uint8x16_t) + +/* +** test_vluti2q_lane_f16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_f16, float16x8_t, float16x8_t, uint8x8_t) + +/* +** test_vluti2q_laneq_f16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_f16, float16x8_t, float16x8_t, uint8x16_t) + +/* +** test_vluti2_lane_bf16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_bf16, bfloat16x8_t, bfloat16x4_t, uint8x8_t) + +/* +** test_vluti2_laneq_bf16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_bf16, bfloat16x8_t, bfloat16x4_t, uint8x16_t) + +/* +** test_vluti2q_lane_bf16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_bf16, bfloat16x8_t, bfloat16x8_t, uint8x8_t) + +/* +** test_vluti2q_laneq_bf16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_bf16, bfloat16x8_t, bfloat16x8_t, uint8x16_t) + +/* +** test_vluti2_lane_p16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_lane_p16, poly16x8_t, poly16x4_t, uint8x8_t) + +/* +** test_vluti2_laneq_p16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2_laneq_p16, poly16x8_t, poly16x4_t, uint8x16_t) + +/* +** test_vluti2q_lane_p16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_lane_p16, poly16x8_t, poly16x8_t, uint8x8_t) + +/* +** test_vluti2q_laneq_p16: +** luti2 v[0-9]+\.8h, {v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI2 (vluti2q_laneq_p16, poly16x8_t, poly16x8_t, uint8x16_t) diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vluti4.c b/gcc/testsuite/gcc.target/aarch64/neon/vluti4.c new file mode 100644 index 000000000000..99ff77ae4f6b --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vluti4.c @@ -0,0 +1,138 @@ +/* { dg-do compile } */ +/* { dg-additional-options "-O3 -march=armv9-a+lut" } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +#define TEST_LUTI4(NAME, RET_TYPE, ARG_1_TYPE, ARG_2_TYPE) \ + RET_TYPE test_##NAME (ARG_1_TYPE vn, ARG_2_TYPE vm) \ + { \ + return NAME (vn, vm, 0); \ + } + +/* +** test_vluti4q_lane_u8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_u8, uint8x16_t, uint8x16_t, uint8x8_t) + +/* +** test_vluti4q_laneq_u8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_u8, uint8x16_t, uint8x16_t, uint8x16_t) + +/* +** test_vluti4q_lane_s8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_s8, int8x16_t, int8x16_t, uint8x8_t) + +/* +** test_vluti4q_laneq_s8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_s8, int8x16_t, int8x16_t, uint8x16_t) + +/* +** test_vluti4q_lane_p8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_p8, poly8x16_t, poly8x16_t, uint8x8_t) + +/* +** test_vluti4q_laneq_p8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_p8, poly8x16_t, poly8x16_t, uint8x16_t) + +/* +** test_vluti4q_lane_mf8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_mf8, mfloat8x16_t, mfloat8x16_t, uint8x8_t) + +/* +** test_vluti4q_laneq_mf8: +** luti4 v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_mf8, mfloat8x16_t, mfloat8x16_t, uint8x16_t) + +/* +** test_vluti4q_lane_u16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_u16_x2, uint16x8_t, uint16x8x2_t, uint8x8_t) + +/* +** test_vluti4q_laneq_u16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_u16_x2, uint16x8_t, uint16x8x2_t, uint8x16_t) + +/* +** test_vluti4q_lane_s16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_s16_x2, int16x8_t, int16x8x2_t, uint8x8_t) + +/* +** test_vluti4q_laneq_s16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_s16_x2, int16x8_t, int16x8x2_t, uint8x16_t) + +/* +** test_vluti4q_lane_f16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_f16_x2, float16x8_t, float16x8x2_t, uint8x8_t) + +/* +** test_vluti4q_laneq_f16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_f16_x2, float16x8_t, float16x8x2_t, uint8x16_t) + +/* +** test_vluti4q_lane_bf16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_bf16_x2, bfloat16x8_t, bfloat16x8x2_t, uint8x8_t) + +/* +** test_vluti4q_laneq_bf16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_bf16_x2, bfloat16x8_t, bfloat16x8x2_t, + uint8x16_t) + +/* +** test_vluti4q_lane_p16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_lane_p16_x2, poly16x8_t, poly16x8x2_t, uint8x8_t) + +/* +** test_vluti4q_laneq_p16_x2: +** luti4 v[0-9]+\.8h, {v[0-9]+\.8h, v[0-9]+\.8h}, v[0-9]+\[0\] +** ret +*/ +TEST_LUTI4 (vluti4q_laneq_p16_x2, poly16x8_t, poly16x8x2_t, uint8x16_t) diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vtbl.c b/gcc/testsuite/gcc.target/aarch64/neon/vtbl.c new file mode 100644 index 000000000000..cece43af692f --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vtbl.c @@ -0,0 +1,376 @@ +/* { dg-do compile } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +/* +** test_vtbl1_s8: +** fmov d[0-9]+, d[0-9]+ +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl1_s8, int8x8_t, int8x8_t, int8x8_t) + +/* +** test_vtbl1_u8: +** fmov d[0-9]+, d[0-9]+ +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl1_u8, uint8x8_t, uint8x8_t, uint8x8_t) + +/* +** test_vtbl1_p8: +** fmov d[0-9]+, d[0-9]+ +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl1_p8, poly8x8_t, poly8x8_t, uint8x8_t) + +/* +** test_vtbl1_mf8: +** fmov d[0-9]+, d[0-9]+ +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl1_mf8, mfloat8x8_t, mfloat8x8_t, uint8x8_t) + +/* +** test_vtbl2_s8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl2_s8, int8x8_t, int8x8x2_t, int8x8_t) + +/* +** test_vtbl2_u8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl2_u8, uint8x8_t, uint8x8x2_t, uint8x8_t) + +/* +** test_vtbl2_p8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl2_p8, poly8x8_t, poly8x8x2_t, uint8x8_t) + +/* +** test_vtbl2_mf8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl2_mf8, mfloat8x8_t, mfloat8x8x2_t, uint8x8_t) + +/* +** test_vtbl3_s8: +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** fmov d([0-9]+), d[0-9]+ +** tbl v[0-9]+\.8b, {v\1\.16b - v\2\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl3_s8, int8x8_t, int8x8x3_t, int8x8_t) + +/* +** test_vtbl3_u8: +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** fmov d([0-9]+), d[0-9]+ +** tbl v[0-9]+\.8b, {v\1\.16b - v\2\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl3_u8, uint8x8_t, uint8x8x3_t, uint8x8_t) + +/* +** test_vtbl3_p8: +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** fmov d([0-9]+), d[0-9]+ +** tbl v[0-9]+\.8b, {v\1\.16b - v\2\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl3_p8, poly8x8_t, poly8x8x3_t, uint8x8_t) + +/* +** test_vtbl3_mf8: +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** fmov d([0-9]+), d[0-9]+ +** tbl v[0-9]+\.8b, {v\1\.16b - v\2\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl3_mf8, mfloat8x8_t, mfloat8x8x3_t, uint8x8_t) + +/* +** test_vtbl4_s8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl4_s8, int8x8_t, int8x8x4_t, int8x8_t) + +/* +** test_vtbl4_u8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl4_u8, uint8x8_t, uint8x8x4_t, uint8x8_t) + +/* +** test_vtbl4_p8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl4_p8, poly8x8_t, poly8x8x4_t, uint8x8_t) + +/* +** test_vtbl4_mf8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vtbl4_mf8, mfloat8x8_t, mfloat8x8x4_t, uint8x8_t) + +/* +** test_vqtbl1_s8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl1_s8, int8x8_t, int8x16_t, uint8x8_t) + +/* +** test_vqtbl1_u8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl1_u8, uint8x8_t, uint8x16_t, uint8x8_t) + +/* +** test_vqtbl1_p8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl1_p8, poly8x8_t, poly8x16_t, uint8x8_t) + +/* +** test_vqtbl1_mf8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl1_mf8, mfloat8x8_t, mfloat8x16_t, uint8x8_t) + +/* +** test_vqtbl2_s8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl2_s8, int8x8_t, int8x16x2_t, uint8x8_t) + +/* +** test_vqtbl2_u8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl2_u8, uint8x8_t, uint8x16x2_t, uint8x8_t) + +/* +** test_vqtbl2_p8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl2_p8, poly8x8_t, poly8x16x2_t, uint8x8_t) + +/* +** test_vqtbl2_mf8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl2_mf8, mfloat8x8_t, mfloat8x16x2_t, uint8x8_t) + +/* +** test_vqtbl3_s8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl3_s8, int8x8_t, int8x16x3_t, uint8x8_t) + +/* +** test_vqtbl3_u8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl3_u8, uint8x8_t, uint8x16x3_t, uint8x8_t) + +/* +** test_vqtbl3_p8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl3_p8, poly8x8_t, poly8x16x3_t, uint8x8_t) + +/* +** test_vqtbl3_mf8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl3_mf8, mfloat8x8_t, mfloat8x16x3_t, uint8x8_t) + +/* +** test_vqtbl4_s8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl4_s8, int8x8_t, int8x16x4_t, uint8x8_t) + +/* +** test_vqtbl4_u8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl4_u8, uint8x8_t, uint8x16x4_t, uint8x8_t) + +/* +** test_vqtbl4_p8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl4_p8, poly8x8_t, poly8x16x4_t, uint8x8_t) + +/* +** test_vqtbl4_mf8: +** tbl v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_BINARY (vqtbl4_mf8, mfloat8x8_t, mfloat8x16x4_t, uint8x8_t) + +/* +** test_vqtbl1q_s8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl1q_s8, int8x16_t, int8x16_t, uint8x16_t) + +/* +** test_vqtbl1q_u8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl1q_u8, uint8x16_t, uint8x16_t, uint8x16_t) + +/* +** test_vqtbl1q_p8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl1q_p8, poly8x16_t, poly8x16_t, uint8x16_t) + +/* +** test_vqtbl1q_mf8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl1q_mf8, mfloat8x16_t, mfloat8x16_t, uint8x16_t) + +/* +** test_vqtbl2q_s8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl2q_s8, int8x16_t, int8x16x2_t, uint8x16_t) + +/* +** test_vqtbl2q_u8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl2q_u8, uint8x16_t, uint8x16x2_t, uint8x16_t) + +/* +** test_vqtbl2q_p8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl2q_p8, poly8x16_t, poly8x16x2_t, uint8x16_t) + +/* +** test_vqtbl2q_mf8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl2q_mf8, mfloat8x16_t, mfloat8x16x2_t, uint8x16_t) + +/* +** test_vqtbl3q_s8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl3q_s8, int8x16_t, int8x16x3_t, uint8x16_t) + +/* +** test_vqtbl3q_u8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl3q_u8, uint8x16_t, uint8x16x3_t, uint8x16_t) + +/* +** test_vqtbl3q_p8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl3q_p8, poly8x16_t, poly8x16x3_t, uint8x16_t) + +/* +** test_vqtbl3q_mf8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl3q_mf8, mfloat8x16_t, mfloat8x16x3_t, uint8x16_t) + +/* +** test_vqtbl4q_s8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl4q_s8, int8x16_t, int8x16x4_t, uint8x16_t) + +/* +** test_vqtbl4q_u8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl4q_u8, uint8x16_t, uint8x16x4_t, uint8x16_t) + +/* +** test_vqtbl4q_p8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl4q_p8, poly8x16_t, poly8x16x4_t, uint8x16_t) + +/* +** test_vqtbl4q_mf8: +** tbl v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_BINARY (vqtbl4q_mf8, mfloat8x16_t, mfloat8x16x4_t, uint8x16_t) diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vtbx.c b/gcc/testsuite/gcc.target/aarch64/neon/vtbx.c new file mode 100644 index 000000000000..3bdf78170232 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vtbx.c @@ -0,0 +1,405 @@ +/* { dg-do compile } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +/* +** test_vtbx1_s8: +** fmov d[0-9]+, d[0-9]+ +** movi v[0-9]+\.8b, 0x7 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** cmhs v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx1_s8, int8x8_t, int8x8_t, int8x8_t, int8x8_t) + +/* +** test_vtbx1_u8: +** fmov d[0-9]+, d[0-9]+ +** movi v[0-9]+\.8b, 0x7 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** cmhs v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx1_u8, uint8x8_t, uint8x8_t, uint8x8_t, uint8x8_t) + +/* +** test_vtbx1_p8: +** fmov d[0-9]+, d[0-9]+ +** movi v[0-9]+\.8b, 0x7 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** cmhs v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx1_p8, poly8x8_t, poly8x8_t, poly8x8_t, uint8x8_t) + +/* +** test_vtbx1_mf8: +** fmov d[0-9]+, d[0-9]+ +** movi v[0-9]+\.8b, 0x7 +** tbl v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** cmhs v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v[0-9]+\.8b, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx1_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x8_t, uint8x8_t) + + +/* +** test_vtbx2_s8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx2_s8, int8x8_t, int8x8_t, int8x8x2_t, int8x8_t) + +/* +** test_vtbx2_u8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx2_u8, uint8x8_t, uint8x8_t, uint8x8x2_t, uint8x8_t) + +/* +** test_vtbx2_p8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx2_p8, poly8x8_t, poly8x8_t, poly8x8x2_t, uint8x8_t) + +/* +** test_vtbx2_mf8: +** sub sp, sp, #16 +** stp d[0-9]+, d[0-9]+, \[sp\] +** ldr q[0-9]+, \[sp\] +** add sp, sp, 16 +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx2_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x8x2_t, uint8x8_t) + +/* +** test_vtbx3_s8: +** fmov d([0-9]+), d[0-9]+ +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v\2\.8b, {v\2\.16b - v\1\.16b}, v[0-9]+\.8b +** movi v\1\.8b, 0x17 +** cmhs v\1\.8b, v\1\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v\2\.8b, v\1\.8b +** ret +*/ +TEST_TERNARY (vtbx3_s8, int8x8_t, int8x8_t, int8x8x3_t, int8x8_t) + +/* +** test_vtbx3_u8: +** fmov d([0-9]+), d[0-9]+ +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v\2\.8b, {v\2\.16b - v\1\.16b}, v[0-9]+\.8b +** movi v\1\.8b, 0x17 +** cmhs v\1\.8b, v\1\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v\2\.8b, v\1\.8b +** ret +*/ +TEST_TERNARY (vtbx3_u8, uint8x8_t, uint8x8_t, uint8x8x3_t, uint8x8_t) + +/* +** test_vtbx3_p8: +** fmov d([0-9]+), d[0-9]+ +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v\2\.8b, {v\2\.16b - v\1\.16b}, v[0-9]+\.8b +** movi v\1\.8b, 0x17 +** cmhs v\1\.8b, v\1\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v\2\.8b, v\1\.8b +** ret +*/ +TEST_TERNARY (vtbx3_p8, poly8x8_t, poly8x8_t, poly8x8x3_t, uint8x8_t) + +/* +** test_vtbx3_mf8: +** fmov d([0-9]+), d[0-9]+ +** zip1 v([0-9]+)\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbl v\2\.8b, {v\2\.16b - v\1\.16b}, v[0-9]+\.8b +** movi v\1\.8b, 0x17 +** cmhs v\1\.8b, v\1\.8b, v[0-9]+\.8b +** bit v[0-9]+\.8b, v\2\.8b, v\1\.8b +** ret +*/ +TEST_TERNARY (vtbx3_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x8x3_t, uint8x8_t) + +/* +** test_vtbx4_s8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx4_s8, int8x8_t, int8x8_t, int8x8x4_t, int8x8_t) + +/* +** test_vtbx4_u8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx4_u8, uint8x8_t, uint8x8_t, uint8x8x4_t, uint8x8_t) + +/* +** test_vtbx4_p8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx4_p8, poly8x8_t, poly8x8_t, poly8x8x4_t, uint8x8_t) + +/* +** test_vtbx4_mf8: +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** zip1 v[0-9]+\.2d, v[0-9]+\.2d, v[0-9]+\.2d +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vtbx4_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x8x4_t, uint8x8_t) + +/* +** test_vqtbx1_s8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx1_s8, int8x8_t, int8x8_t, int8x16_t, uint8x8_t) + +/* +** test_vqtbx1_u8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx1_u8, uint8x8_t, uint8x8_t, uint8x16_t, uint8x8_t) + +/* +** test_vqtbx1_p8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx1_p8, poly8x8_t, poly8x8_t, poly8x16_t, uint8x8_t) + +/* +** test_vqtbx1_mf8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx1_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x16_t, + uint8x8_t) + +/* +** test_vqtbx2_s8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx2_s8, int8x8_t, int8x8_t, int8x16x2_t, uint8x8_t) + +/* +** test_vqtbx2_u8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx2_u8, uint8x8_t, uint8x8_t, uint8x16x2_t, uint8x8_t) + +/* +** test_vqtbx2_p8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx2_p8, poly8x8_t, poly8x8_t, poly8x16x2_t, uint8x8_t) + +/* +** test_vqtbx2_mf8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx2_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x16x2_t, + uint8x8_t) + +/* +** test_vqtbx3_s8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx3_s8, int8x8_t, int8x8_t, int8x16x3_t, uint8x8_t) + +/* +** test_vqtbx3_u8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx3_u8, uint8x8_t, uint8x8_t, uint8x16x3_t, uint8x8_t) + +/* +** test_vqtbx3_p8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx3_p8, poly8x8_t, poly8x8_t, poly8x16x3_t, uint8x8_t) + +/* +** test_vqtbx3_mf8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx3_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x16x3_t, + uint8x8_t) + +/* +** test_vqtbx4_s8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx4_s8, int8x8_t, int8x8_t, int8x16x4_t, uint8x8_t) + +/* +** test_vqtbx4_u8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx4_u8, uint8x8_t, uint8x8_t, uint8x16x4_t, uint8x8_t) + +/* +** test_vqtbx4_p8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx4_p8, poly8x8_t, poly8x8_t, poly8x16x4_t, uint8x8_t) + +/* +** test_vqtbx4_mf8: +** tbx v[0-9]+\.8b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.8b +** ret +*/ +TEST_TERNARY (vqtbx4_mf8, mfloat8x8_t, mfloat8x8_t, mfloat8x16x4_t, + uint8x8_t) + +/* +** test_vqtbx1q_s8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx1q_s8, int8x16_t, int8x16_t, int8x16_t, uint8x16_t) + +/* +** test_vqtbx1q_u8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx1q_u8, uint8x16_t, uint8x16_t, uint8x16_t, uint8x16_t) + +/* +** test_vqtbx1q_p8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx1q_p8, poly8x16_t, poly8x16_t, poly8x16_t, uint8x16_t) + +/* +** test_vqtbx1q_mf8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx1q_mf8, mfloat8x16_t, mfloat8x16_t, mfloat8x16_t, uint8x16_t) + +/* +** test_vqtbx2q_s8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx2q_s8, int8x16_t, int8x16_t, int8x16x2_t, uint8x16_t) + +/* +** test_vqtbx2q_u8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx2q_u8, uint8x16_t, uint8x16_t, uint8x16x2_t, uint8x16_t) + +/* +** test_vqtbx2q_p8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx2q_p8, poly8x16_t, poly8x16_t, poly8x16x2_t, uint8x16_t) + +/* +** test_vqtbx2q_mf8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx2q_mf8, mfloat8x16_t, mfloat8x16_t, mfloat8x16x2_t, uint8x16_t) + +/* +** test_vqtbx3q_s8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx3q_s8, int8x16_t, int8x16_t, int8x16x3_t, uint8x16_t) + +/* +** test_vqtbx3q_u8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx3q_u8, uint8x16_t, uint8x16_t, uint8x16x3_t, uint8x16_t) + +/* +** test_vqtbx3q_p8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx3q_p8, poly8x16_t, poly8x16_t, poly8x16x3_t, uint8x16_t) + +/* +** test_vqtbx3q_mf8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx3q_mf8, mfloat8x16_t, mfloat8x16_t, mfloat8x16x3_t, uint8x16_t) + +/* +** test_vqtbx4q_s8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx4q_s8, int8x16_t, int8x16_t, int8x16x4_t, uint8x16_t) + +/* +** test_vqtbx4q_u8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx4q_u8, uint8x16_t, uint8x16_t, uint8x16x4_t, uint8x16_t) + +/* +** test_vqtbx4q_p8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx4q_p8, poly8x16_t, poly8x16_t, poly8x16x4_t, uint8x16_t) + +/* +** test_vqtbx4q_mf8: +** tbx v[0-9]+\.16b, {v[0-9]+\.16b - v[0-9]+\.16b}, v[0-9]+\.16b +** ret +*/ +TEST_TERNARY (vqtbx4q_mf8, mfloat8x16_t, mfloat8x16_t, mfloat8x16x4_t, uint8x16_t)
