From 683e93d1975f676d00096b7c93b942dbd0491800 Mon Sep 17 00:00:00 2001 From: Richard Sandiford Date: Tue, 7 Jan 2020 10:18:14 +0000 Subject: [PATCH] [AArch64] Use type attributes to mark types that use the SVE PCS The SVE port needs to maintain a different type identity for GNU vectors and "SVE vectors", since the types use different ABIs. Until now we've done that using pointer equality between the TYPE_MAIN_VARIANT and the built-in SVE type. However, as Richard B noted, that doesn't work well for LTO, where we stream both GNU and SVE types from a file instead of creating them directly. We need a mechanism for distinguishing the types using streamed type information. This patch does that using a new type attribute. This attribute is only meant to be used for the built-in SVE types and shouldn't be user-visible. The patch tries to ensure this by including a space in the attribute name, like we already do for things like "fn spec" and "omp declare simd". 2020-01-07 Richard Sandiford gcc/ * config/aarch64/aarch64-protos.h (aarch64_sve::svbool_type_p) (aarch64_sve::nvectors_if_data_type): Replace with... (aarch64_sve::builtin_type_p): ...this. * config/aarch64/aarch64-sve-builtins.cc: Include attribs.h. (find_vector_type): Delete. (add_sve_type_attribute): New function. (lookup_sve_type_attribute): Likewise. (register_builtin_types): Add an "SVE type" attribute to each type. (register_tuple_type): Likewise. (svbool_type_p, nvectors_if_data_type): Delete. (mangle_builtin_type): Use lookup_sve_type_attribute. (builtin_type_p): Likewise. Add an overload that returns the number of constituent vector and predicate registers. * config/aarch64/aarch64.c (aarch64_sve_argument_p): Delete. (aarch64_returns_value_in_sve_regs_p): Use aarch64_sve::builtin_type_p instead of aarch64_sve_argument_p. (aarch64_takes_arguments_in_sve_regs_p): Likewise. (aarch64_pass_by_reference): Likewise. (aarch64_function_value_1): Likewise. (aarch64_return_in_memory): Likewise. (aarch64_layout_arg): Likewise. gcc/testsuite/ * g++.target/aarch64/sve/acle/general-c++/mangle_5.C: New test. * gcc.target/aarch64/sve/pcs/asm_1.c: Likewise. * gcc.target/aarch64/sve/pcs/asm_2.c: Likewise. * gcc.target/aarch64/sve/pcs/asm_3.c: Likewise. From-SVN: r279953 --- gcc/ChangeLog | 24 +++++ gcc/config/aarch64/aarch64-protos.h | 3 +- gcc/config/aarch64/aarch64-sve-builtins.cc | 113 +++++++++++---------- gcc/config/aarch64/aarch64.c | 36 ++----- gcc/testsuite/ChangeLog | 7 ++ .../aarch64/sve/acle/general-c++/mangle_5.C | 8 ++ gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_1.c | 70 +++++++++++++ gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_2.c | 4 + gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_3.c | 68 +++++++++++++ 9 files changed, 246 insertions(+), 87 deletions(-) create mode 100644 gcc/testsuite/g++.target/aarch64/sve/acle/general-c++/mangle_5.C create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_1.c create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_2.c create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_3.c diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 0ead18a..3aa2061 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,27 @@ +2020-01-07 Richard Sandiford + + * config/aarch64/aarch64-protos.h (aarch64_sve::svbool_type_p) + (aarch64_sve::nvectors_if_data_type): Replace with... + (aarch64_sve::builtin_type_p): ...this. + * config/aarch64/aarch64-sve-builtins.cc: Include attribs.h. + (find_vector_type): Delete. + (add_sve_type_attribute): New function. + (lookup_sve_type_attribute): Likewise. + (register_builtin_types): Add an "SVE type" attribute to each type. + (register_tuple_type): Likewise. + (svbool_type_p, nvectors_if_data_type): Delete. + (mangle_builtin_type): Use lookup_sve_type_attribute. + (builtin_type_p): Likewise. Add an overload that returns the + number of constituent vector and predicate registers. + * config/aarch64/aarch64.c (aarch64_sve_argument_p): Delete. + (aarch64_returns_value_in_sve_regs_p): Use aarch64_sve::builtin_type_p + instead of aarch64_sve_argument_p. + (aarch64_takes_arguments_in_sve_regs_p): Likewise. + (aarch64_pass_by_reference): Likewise. + (aarch64_function_value_1): Likewise. + (aarch64_return_in_memory): Likewise. + (aarch64_layout_arg): Likewise. + 2020-01-07 Jakub Jelinek PR tree-optimization/93156 diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h index bedae23..a400430 100644 --- a/gcc/config/aarch64/aarch64-protos.h +++ b/gcc/config/aarch64/aarch64-protos.h @@ -706,8 +706,7 @@ namespace aarch64_sve { void handle_arm_sve_h (); tree builtin_decl (unsigned, bool); bool builtin_type_p (const_tree); - bool svbool_type_p (const_tree); - unsigned int nvectors_if_data_type (const_tree); + bool builtin_type_p (const_tree, unsigned int *, unsigned int *); const char *mangle_builtin_type (const_tree); tree resolve_overloaded_builtin (location_t, unsigned int, vec *); diff --git a/gcc/config/aarch64/aarch64-sve-builtins.cc b/gcc/config/aarch64/aarch64-sve-builtins.cc index 6598218..566ebae 100644 --- a/gcc/config/aarch64/aarch64-sve-builtins.cc +++ b/gcc/config/aarch64/aarch64-sve-builtins.cc @@ -47,6 +47,7 @@ #include "gimple-fold.h" #include "langhooks.h" #include "stringpool.h" +#include "attribs.h" #include "aarch64-sve-builtins.h" #include "aarch64-sve-builtins-base.h" #include "aarch64-sve-builtins-shapes.h" @@ -418,18 +419,31 @@ static hash_table *function_table; when the required extension is disabled. */ static bool reported_missing_extension_p; -/* If TYPE is an ACLE vector type, return the associated vector_type, - otherwise return NUM_VECTOR_TYPES. */ -static vector_type_index -find_vector_type (const_tree type) +/* Record that TYPE is an ABI-defined SVE type that contains NUM_ZR SVE vectors + and NUM_PR SVE predicates. MANGLED_NAME, if nonnull, is the ABI-defined + mangling of the type. */ +static void +add_sve_type_attribute (tree type, unsigned int num_zr, unsigned int num_pr, + const char *mangled_name) { - /* A linear search should be OK here, since the code isn't hot and - the number of types is only small. */ - type = TYPE_MAIN_VARIANT (type); - for (unsigned int i = 0; i < NUM_VECTOR_TYPES; ++i) - if (type == abi_vector_types[i]) - return vector_type_index (i); - return NUM_VECTOR_TYPES; + tree mangled_name_tree + = (mangled_name ? get_identifier (mangled_name) : NULL_TREE); + + tree value = tree_cons (NULL_TREE, mangled_name_tree, NULL_TREE); + value = tree_cons (NULL_TREE, size_int (num_pr), value); + value = tree_cons (NULL_TREE, size_int (num_zr), value); + TYPE_ATTRIBUTES (type) = tree_cons (get_identifier ("SVE type"), value, + TYPE_ATTRIBUTES (type)); +} + +/* If TYPE is an ABI-defined SVE type, return its attribute descriptor, + otherwise return null. */ +static tree +lookup_sve_type_attribute (const_tree type) +{ + if (type == error_mark_node) + return NULL_TREE; + return lookup_attribute ("SVE type", TYPE_ATTRIBUTES (type)); } /* If TYPE is a valid SVE element type, return the corresponding type @@ -2986,6 +3000,7 @@ register_builtin_types () { tree eltype = scalar_types[i]; tree vectype; + unsigned int num_zr = 0, num_pr = 0; if (eltype == boolean_type_node) { vectype = build_truth_vector_type_for_mode (BYTES_PER_SVE_VECTOR, @@ -2995,6 +3010,7 @@ register_builtin_types () && TYPE_ALIGN (vectype) == 16 && known_eq (wi::to_poly_offset (TYPE_SIZE (vectype)), BYTES_PER_SVE_VECTOR)); + num_pr = 1; } else { @@ -3006,12 +3022,15 @@ register_builtin_types () && TYPE_ALIGN (vectype) == 128 && known_eq (wi::to_poly_offset (TYPE_SIZE (vectype)), BITS_PER_SVE_VECTOR)); + num_zr = 1; } vectype = build_distinct_type_copy (vectype); gcc_assert (vectype == TYPE_MAIN_VARIANT (vectype)); SET_TYPE_STRUCTURAL_EQUALITY (vectype); TYPE_ARTIFICIAL (vectype) = 1; TYPE_INDIVISIBLE_P (vectype) = 1; + add_sve_type_attribute (vectype, num_zr, num_pr, + vector_types[i].mangled_name); abi_vector_types[i] = vectype; lang_hooks.types.register_builtin_type (vectype, vector_types[i].abi_name); @@ -3076,6 +3095,7 @@ register_tuple_type (unsigned int num_vectors, vector_type_index type) get_identifier ("__val"), array_type); DECL_FIELD_CONTEXT (field) = tuple_type; TYPE_FIELDS (tuple_type) = field; + add_sve_type_attribute (tuple_type, num_vectors, 0, NULL); layout_type (tuple_type); gcc_assert (VECTOR_MODE_P (TYPE_MODE (tuple_type)) && TYPE_MODE_RAW (tuple_type) == TYPE_MODE (tuple_type) @@ -3247,64 +3267,45 @@ expand_builtin (unsigned int code, tree exp, rtx target) return function_expander (rfn.instance, rfn.decl, exp, target).expand (); } -/* Return true if TYPE is the ABI-defined __SVBool_t type. */ -bool -svbool_type_p (const_tree type) -{ - tree abi_type = abi_vector_types[VECTOR_TYPE_svbool_t]; - return type != error_mark_node && TYPE_MAIN_VARIANT (type) == abi_type; -} - /* If TYPE is a built-in type defined by the SVE ABI, return the mangled name, otherwise return NULL. */ const char * mangle_builtin_type (const_tree type) { - if (type == error_mark_node) - return NULL; - - vector_type_index vtype = find_vector_type (type); - if (vtype != NUM_VECTOR_TYPES) - return vector_types[vtype].mangled_name; - + /* ??? The C++ frontend normally strips qualifiers and attributes before + calling this hook, adding separate mangling for attributes that affect + type identity. Fortunately the type copy will have the same TYPE_NAME + as the original, so we can get the attributes from there. */ + if (TYPE_NAME (type) && TREE_CODE (TYPE_NAME (type)) == TYPE_DECL) + type = TREE_TYPE (TYPE_NAME (type)); + if (tree attr = lookup_sve_type_attribute (type)) + if (tree id = TREE_VALUE (chain_index (2, TREE_VALUE (attr)))) + return IDENTIFIER_POINTER (id); return NULL; } -/* If TYPE is one of the ABI-defined SVE vector types, or an ACLE-defined - tuple of them, return the number of vectors it contains. Return 0 - otherwise. */ -unsigned int -nvectors_if_data_type (const_tree type) +/* Return true if TYPE is a built-in SVE type defined by the ABI or ACLE. */ +bool +builtin_type_p (const_tree type) { - if (type == error_mark_node) - return 0; - - type = TYPE_MAIN_VARIANT (type); - if (VECTOR_TYPE_P (type)) - { - vector_type_index type_id = find_vector_type (type); - if (type_id != VECTOR_TYPE_svbool_t && type_id != NUM_VECTOR_TYPES) - return 1; - } - else if (TREE_CODE (type) == RECORD_TYPE) - { - for (unsigned int size_i = 1; size_i < MAX_TUPLE_SIZE; ++size_i) - for (unsigned int type_i = 0; type_i < NUM_VECTOR_TYPES; ++type_i) - { - tree tuple_type = acle_vector_types[size_i][type_i]; - if (tuple_type && type == TYPE_MAIN_VARIANT (tuple_type)) - return size_i + 1; - } - } - - return 0; + return lookup_sve_type_attribute (type); } -/* Return true if TYPE is a built-in type defined by the SVE ABI. */ +/* Return true if TYPE is a built-in SVE type defined by the ABI or ACLE. + If so, store the number of constituent SVE vectors in *NUM_ZR and the + number of constituent SVE predicates in *NUM_PR. */ bool -builtin_type_p (const_tree type) +builtin_type_p (const_tree type, unsigned int *num_zr, unsigned int *num_pr) { - return svbool_type_p (type) || nvectors_if_data_type (type) > 0; + if (tree attr = lookup_sve_type_attribute (type)) + { + tree num_zr_node = TREE_VALUE (attr); + tree num_pr_node = TREE_CHAIN (num_zr_node); + *num_zr = tree_to_uhwi (TREE_VALUE (num_zr_node)); + *num_pr = tree_to_uhwi (TREE_VALUE (num_pr_node)); + return true; + } + return false; } /* Implement TARGET_VERIFY_TYPE_CONTEXT for SVE types. */ diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c index 85cadef..66e20be 100644 --- a/gcc/config/aarch64/aarch64.c +++ b/gcc/config/aarch64/aarch64.c @@ -1246,6 +1246,7 @@ static const struct attribute_spec aarch64_attribute_table[] = affects_type_identity, handler, exclude } */ { "aarch64_vector_pcs", 0, 0, false, true, true, true, handle_aarch64_vector_pcs_attribute, NULL }, + { "SVE type", 3, 3, false, true, false, true, NULL, NULL }, { NULL, 0, 0, false, false, false, false, NULL, NULL } }; @@ -2042,37 +2043,15 @@ aarch64_hard_regno_mode_ok (unsigned regno, machine_mode mode) true, set *NUM_ZR and *NUM_PR to the number of required Z and P registers respectively. */ -static bool -aarch64_sve_argument_p (const_tree type, unsigned int *num_zr, - unsigned int *num_pr) -{ - if (aarch64_sve::svbool_type_p (type)) - { - *num_pr = 1; - *num_zr = 0; - return true; - } - - if (unsigned int nvectors = aarch64_sve::nvectors_if_data_type (type)) - { - *num_pr = 0; - *num_zr = nvectors; - return true; - } - - return false; -} - /* Return true if a function with type FNTYPE returns its value in SVE vector or predicate registers. */ static bool aarch64_returns_value_in_sve_regs_p (const_tree fntype) { - unsigned int num_zr, num_pr; tree return_type = TREE_TYPE (fntype); return (return_type != error_mark_node - && aarch64_sve_argument_p (return_type, &num_zr, &num_pr)); + && aarch64_sve::builtin_type_p (return_type)); } /* Return true if a function with type FNTYPE takes arguments in @@ -2096,8 +2075,7 @@ aarch64_takes_arguments_in_sve_regs_p (const_tree fntype) function_arg_info arg (arg_type, /*named=*/true); apply_pass_by_reference_rules (&args_so_far_v, arg); - unsigned int num_zr, num_pr; - if (aarch64_sve_argument_p (arg.type, &num_zr, &num_pr)) + if (aarch64_sve::builtin_type_p (arg.type)) return true; targetm.calls.function_arg_advance (args_so_far, arg); @@ -4876,7 +4854,7 @@ aarch64_pass_by_reference (cumulative_args_t pcum_v, int nregs; unsigned int num_zr, num_pr; - if (arg.type && aarch64_sve_argument_p (arg.type, &num_zr, &num_pr)) + if (arg.type && aarch64_sve::builtin_type_p (arg.type, &num_zr, &num_pr)) { if (pcum && !pcum->silent_p && !TARGET_SVE) /* We can't gracefully recover at this point, so make this a @@ -4955,7 +4933,7 @@ static rtx aarch64_function_value_1 (const_tree type, machine_mode mode) { unsigned int num_zr, num_pr; - if (type && aarch64_sve_argument_p (type, &num_zr, &num_pr)) + if (type && aarch64_sve::builtin_type_p (type, &num_zr, &num_pr)) { /* Don't raise an error here if we're called when SVE is disabled, since this is really just a query function. Other code must @@ -5098,7 +5076,7 @@ aarch64_return_in_memory (const_tree type, const_tree fndecl ATTRIBUTE_UNUSED) return false; unsigned int num_zr, num_pr; - if (type && aarch64_sve_argument_p (type, &num_zr, &num_pr)) + if (type && aarch64_sve::builtin_type_p (type, &num_zr, &num_pr)) { /* All SVE types we support fit in registers. For example, it isn't yet possible to define an aggregate of 9+ SVE vectors or 5+ SVE @@ -5226,7 +5204,7 @@ aarch64_layout_arg (cumulative_args_t pcum_v, const function_arg_info &arg, pcum->aapcs_arg_processed = true; unsigned int num_zr, num_pr; - if (type && aarch64_sve_argument_p (type, &num_zr, &num_pr)) + if (type && aarch64_sve::builtin_type_p (type, &num_zr, &num_pr)) { /* The PCS says that it is invalid to pass an SVE value to an unprototyped function. There is no ABI-defined location we diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index b5c4bc0..9e694fa 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,10 @@ +2020-01-07 Richard Sandiford + + * g++.target/aarch64/sve/acle/general-c++/mangle_5.C: New test. + * gcc.target/aarch64/sve/pcs/asm_1.c: Likewise. + * gcc.target/aarch64/sve/pcs/asm_2.c: Likewise. + * gcc.target/aarch64/sve/pcs/asm_3.c: Likewise. + 2020-01-07 Jakub Jelinek PR tree-optimization/93156 diff --git a/gcc/testsuite/g++.target/aarch64/sve/acle/general-c++/mangle_5.C b/gcc/testsuite/g++.target/aarch64/sve/acle/general-c++/mangle_5.C new file mode 100644 index 0000000..47c1160 --- /dev/null +++ b/gcc/testsuite/g++.target/aarch64/sve/acle/general-c++/mangle_5.C @@ -0,0 +1,8 @@ +typedef const __SVInt8_t foo; +typedef volatile foo bar; + +foo f (foo x) { return x; } +bar g (bar x) { return x; } + +/* { dg-final { scan-assembler {_Z1f10__SVInt8_t:\n} } } */ +/* { dg-final { scan-assembler {_Z1g10__SVInt8_t:\n} } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_1.c b/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_1.c new file mode 100644 index 0000000..8be2094 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_1.c @@ -0,0 +1,70 @@ +/* { dg-do run { target aarch64_sve_hw } } */ +/* { dg-options "-O0 -ffixed-z0 -ffixed-p0" } */ + +#include + +#define ASM_FUNCTION(NAME, RET_TYPE, ARG_TYPE, INSN) \ +extern RET_TYPE NAME (svbool_t, ARG_TYPE); \ + asm( \ +" .type " #NAME ", %function\n" \ +#NAME ":\n" \ +" " INSN "\n" \ +" ret\n" \ +" .size " #NAME ", .-" #NAME "\n" \ +) + +ASM_FUNCTION (u8_callee, uint64_t, svuint8_t, + "uaddv d0, p0, z0.b\n\tfmov x0, d0"); +ASM_FUNCTION (u16_callee, uint64_t, svuint16_t, + "uaddv d0, p0, z0.h\n\tfmov x0, d0"); +ASM_FUNCTION (u32_callee, uint64_t, svuint32_t, + "uaddv d0, p0, z0.s\n\tfmov x0, d0"); +ASM_FUNCTION (u64_callee, uint64_t, svuint64_t, + "uaddv d0, p0, z0.d\n\tfmov x0, d0"); + +ASM_FUNCTION (s8_callee, int64_t, svint8_t, + "saddv d0, p0, z0.b\n\tfmov x0, d0"); +ASM_FUNCTION (s16_callee, int64_t, svint16_t, + "saddv d0, p0, z0.h\n\tfmov x0, d0"); +ASM_FUNCTION (s32_callee, int64_t, svint32_t, + "saddv d0, p0, z0.s\n\tfmov x0, d0"); +ASM_FUNCTION (s64_callee, int64_t, svint64_t, + "uaddv d0, p0, z0.d\n\tfmov x0, d0"); + +ASM_FUNCTION (f16_callee, float16_t, svfloat16_t, "faddv\th0, p0, z0.h"); +ASM_FUNCTION (f32_callee, float32_t, svfloat32_t, "faddv\ts0, p0, z0.s"); +ASM_FUNCTION (f64_callee, float64_t, svfloat64_t, "faddv\td0, p0, z0.d"); + +int +main (void) +{ + if (u8_callee (svptrue_pat_b8 (SV_VL7), svdup_u8 (-1)) != 7 * 0xff) + __builtin_abort (); + if (u16_callee (svptrue_pat_b16 (SV_VL6), svdup_u16 (-1)) != 6 * 0xffff) + __builtin_abort (); + if (u32_callee (svptrue_pat_b32 (SV_VL3), svdup_u32 (-1)) + != 3 * (uint64_t) (uint32_t) -1) + __builtin_abort (); + if (u64_callee (svptrue_pat_b64 (SV_VL2), svdup_u64 ((uint64_t) 1 << 33)) + != (uint64_t) 1 << 34) + __builtin_abort (); + + if (s8_callee (svptrue_pat_b8 (SV_VL7), svdup_s8 (-10)) != -70) + __builtin_abort (); + if (s16_callee (svptrue_pat_b16 (SV_VL6), svdup_s16 (-14)) != -84) + __builtin_abort (); + if (s32_callee (svptrue_pat_b32 (SV_VL3), svdup_s32 (-22)) != -66) + __builtin_abort (); + if (s64_callee (svptrue_pat_b64 (SV_VL2), svdup_s64 ((int64_t) 1 << 33)) + != (int64_t) 1 << 34) + __builtin_abort (); + + if (f16_callee (svptrue_pat_b16 (SV_VL5), svdup_f16 (1.0)) != 5.0) + __builtin_abort (); + if (f32_callee (svptrue_b32 (), svdup_f32 (3.0)) != 3 * svcntw ()) + __builtin_abort (); + if (f64_callee (svptrue_b64 (), svdup_f64 (11.0)) != 11 * svcntd ()) + __builtin_abort (); + + return 0; +} diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_2.c b/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_2.c new file mode 100644 index 0000000..935641b --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_2.c @@ -0,0 +1,4 @@ +/* { dg-do run { target aarch64_sve_hw } } */ +/* { dg-options "-O2 -flto -ffixed-z0 -ffixed-p0" } */ + +#include "asm_1.c" diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_3.c b/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_3.c new file mode 100644 index 0000000..d5a36e5 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/sve/pcs/asm_3.c @@ -0,0 +1,68 @@ +/* { dg-do run { target aarch64_sve_hw } } */ +/* { dg-options "-O0 -ffixed-z0 -ffixed-p0" } */ + +#include + +#define ASM_FUNCTION(NAME, RET_TYPE, ARG_TYPE, INSN) \ +extern RET_TYPE NAME (svbool_t, ARG_TYPE); \ + asm( \ +" .type " #NAME ", %function\n" \ +#NAME ":\n" \ +" " INSN "\n" \ +" ret\n" \ +" .size " #NAME ", .-" #NAME "\n" \ +) + +ASM_FUNCTION (u8_callee, svuint8_t, svuint8x2_t, + "add z0.b, p0/m, z0.b, z1.b"); +ASM_FUNCTION (u16_callee, svuint16_t, svuint16x2_t, + "add z0.h, p0/m, z0.h, z1.h"); +ASM_FUNCTION (u32_callee, svuint32_t, svuint32x2_t, + "add z0.s, p0/m, z0.s, z1.s"); +ASM_FUNCTION (u64_callee, svuint64_t, svuint64x2_t, + "add z0.d, p0/m, z0.d, z1.d"); + +ASM_FUNCTION (s8_callee, svint8_t, svint8x2_t, + "add z0.b, p0/m, z0.b, z1.b"); +ASM_FUNCTION (s16_callee, svint16_t, svint16x2_t, + "add z0.h, p0/m, z0.h, z1.h"); +ASM_FUNCTION (s32_callee, svint32_t, svint32x2_t, + "add z0.s, p0/m, z0.s, z1.s"); +ASM_FUNCTION (s64_callee, svint64_t, svint64x2_t, + "add z0.d, p0/m, z0.d, z1.d"); + +ASM_FUNCTION (f16_callee, svfloat16_t, svfloat16x2_t, + "fadd z0.h, p0/m, z0.h, z1.h"); +ASM_FUNCTION (f32_callee, svfloat32_t, svfloat32x2_t, + "fadd z0.s, p0/m, z0.s, z1.s"); +ASM_FUNCTION (f64_callee, svfloat64_t, svfloat64x2_t, + "fadd z0.d, p0/m, z0.d, z1.d"); + +int +main (void) +{ +#define CHECK(SUFFIX) \ + if (svptest_any (svptrue_b8 (), \ + svcmpne (svptrue_b8 (), \ + SUFFIX##_callee (svptrue_b8 (), \ + svcreate2 (svdup_##SUFFIX (3), \ + svdup_##SUFFIX (6))), \ + svdup_##SUFFIX (9)))) \ + __builtin_abort () + + CHECK (u8); + CHECK (u16); + CHECK (u32); + CHECK (u64); + + CHECK (s8); + CHECK (s16); + CHECK (s32); + CHECK (s64); + + CHECK (f16); + CHECK (f32); + CHECK (f64); + + return 0; +} -- 2.7.4