https://gcc.gnu.org/g:c836fc34d21b6e8545ce66bc9e37c4b44ac5bae5
commit r17-6223-gc836fc34d21b6e8545ce66bc9e37c4b44ac5bae5 Author: Dhruv Chawla <[email protected]> Date: Thu Nov 5 15:43:51 2026 +0530 aarch64: Port NEON shifts by an immediate to the pragma-based framework Port the following families of NEON intrinsics to the pragma-based framework: * vshl_n * vshr_n * vsra_n * vshld_n * vshrd_n * vsrad_n Fix tests where vshr_n by zero is now incorrect. Bootstrapped and regtested on aarch64-linux-gnu. Signed-off-by: Dhruv Chawla <[email protected]> gcc/ChangeLog: * config/aarch64/aarch64-builtins.cc (aarch64_general_gimple_fold_builtin): Remove the ashl, ashr, ashr_simd, lshr and lshr_simd cases which the framework now handles. * config/aarch64/aarch64-neon-builtins-base.cc (vshl_n, vshlq_n, vshld_n, vshr_n, vshrq_n, vshrd_n, vsra_n, vsraq_n, vsrad_n): New function bases. * config/aarch64/aarch64-neon-builtins-base.def: Add the corresponding function group. * config/aarch64/aarch64-simd-builtins.def: Delete the ashl, ashr, ashr_simd, lshr, lshr_simd, ssra_n and usra_n builtins, which no longer have any caller. * config/aarch64/aarch64-simd.md (aarch64_<sra_op>sra_n<mode>, aarch64_<sur>sra_ndi, aarch64_ashr_simddi, aarch64_lshr_simddi): Delete. (aarch64_<sra_op>sra_n<mode>_insn): Rename to ... (*aarch64_<sra_op>sra_n<mode>): ... this, since the expander that called it is gone and combine reaches it by matching. * config/aarch64/aarch64.md (*aarch64_<sra_op>sra_ndi): New pattern. * config/aarch64/iterators.md (UNSPEC_SSRA, UNSPEC_USRA, VSRA): Delete, now that the last pattern using them is gone. (sur): Drop the UNSPEC_SSRA and UNSPEC_USRA entries. * config/aarch64/predicates.md (aarch64_shift_imm64_di): Delete. * config/aarch64/arm_neon.h (vshl_n_s16, vshl_n_s32, vshl_n_s64, vshl_n_s8, vshl_n_u16, vshl_n_u32, vshl_n_u64, vshl_n_u8, vshld_n_s64, vshld_n_u64, vshlq_n_s16, vshlq_n_s32, vshlq_n_s64, vshlq_n_s8, vshlq_n_u16, vshlq_n_u32, vshlq_n_u64, vshlq_n_u8, vshr_n_s16, vshr_n_s32, vshr_n_s64, vshr_n_s8, vshr_n_u16, vshr_n_u32, vshr_n_u64, vshr_n_u8, vshrd_n_s64, vshrd_n_u64, vshrq_n_s16, vshrq_n_s32, vshrq_n_s64, vshrq_n_s8, vshrq_n_u16, vshrq_n_u32, vshrq_n_u64, vshrq_n_u8, vsra_n_s16, vsra_n_s32, vsra_n_s64, vsra_n_s8, vsra_n_u16, vsra_n_u32, vsra_n_u64, vsra_n_u8, vsrad_n_s64, vsrad_n_u64, vsraq_n_s16, vsraq_n_s32, vsraq_n_s64, vsraq_n_s8, vsraq_n_u16, vsraq_n_u32, vsraq_n_u64, vsraq_n_u8): Delete. gcc/testsuite/ChangeLog: * gcc.target/aarch64/neon/vshl_n.c: New test. * gcc.target/aarch64/neon/vshr_n.c: Likewise. * gcc.target/aarch64/neon/vsra_n.c: Likewise. * gcc.target/aarch64/scalar_intrinsics.c: vsrad_n now folds to a shifted add. * gcc.target/aarch64/sshr64_1.c: Drop the shift-by-zero case, which the ACLE does not allow. * gcc.target/aarch64/ushr64_1.c: Likewise. * gcc.target/aarch64/advsimd-intrinsics/vshl-opt-3.c: An arithmetic shift right by the element width now folds. Diff: --- gcc/config/aarch64/aarch64-builtins.cc | 15 - gcc/config/aarch64/aarch64-neon-builtins-base.cc | 5 + gcc/config/aarch64/aarch64-neon-builtins-base.def | 13 + gcc/config/aarch64/aarch64-simd-builtins.def | 9 - gcc/config/aarch64/aarch64-simd.md | 61 +--- gcc/config/aarch64/aarch64.md | 14 + gcc/config/aarch64/arm_neon.h | 382 --------------------- gcc/config/aarch64/iterators.md | 4 - gcc/config/aarch64/predicates.md | 4 - .../aarch64/advsimd-intrinsics/vshl-opt-3.c | 5 +- gcc/testsuite/gcc.target/aarch64/neon/vshl_n.c | 45 +++ gcc/testsuite/gcc.target/aarch64/neon/vshr_n.c | 60 ++++ gcc/testsuite/gcc.target/aarch64/neon/vsra_n.c | 52 +++ .../gcc.target/aarch64/scalar_intrinsics.c | 7 +- gcc/testsuite/gcc.target/aarch64/sshr64_1.c | 21 -- gcc/testsuite/gcc.target/aarch64/ushr64_1.c | 19 - 16 files changed, 197 insertions(+), 519 deletions(-) diff --git a/gcc/config/aarch64/aarch64-builtins.cc b/gcc/config/aarch64/aarch64-builtins.cc index 337ecb1088a6..050d65e86beb 100644 --- a/gcc/config/aarch64/aarch64-builtins.cc +++ b/gcc/config/aarch64/aarch64-builtins.cc @@ -5268,12 +5268,6 @@ aarch64_general_gimple_fold_builtin (unsigned int fcode, gcall *stmt, break; } - BUILTIN_VSDQ_I_DI (BINOP, ashl, 3, DEFAULT) - if (TREE_CODE (args[1]) == INTEGER_CST - && wi::ltu_p (wi::to_wide (args[1]), element_precision (args[0]))) - new_stmt = gimple_build_assign (gimple_call_lhs (stmt), - LSHIFT_EXPR, args[0], args[1]); - break; /* lower saturating add/sub neon builtins to gimple. */ BUILTIN_VSDQ_I (BINOP, ssadd, 3, DEFAULT) BUILTIN_VSDQ_I (BINOPU, usadd, 3, DEFAULT) @@ -5319,15 +5313,6 @@ aarch64_general_gimple_fold_builtin (unsigned int fcode, gcall *stmt, } } break; - BUILTIN_VDQ_I (SHIFTIMM, ashr, 3, DEFAULT) - VAR1 (SHIFTIMM, ashr_simd, 0, DEFAULT, di) - BUILTIN_VDQ_I (USHIFTIMM, lshr, 3, DEFAULT) - VAR1 (USHIFTIMM, lshr_simd, 0, DEFAULT, di) - if (TREE_CODE (args[1]) == INTEGER_CST - && wi::ltu_p (wi::to_wide (args[1]), element_precision (args[0]))) - new_stmt = gimple_build_assign (gimple_call_lhs (stmt), - RSHIFT_EXPR, args[0], args[1]); - break; BUILTIN_GPF (BINOP, fmulx, 0, ALL) { gcc_assert (nargs == 2); diff --git a/gcc/config/aarch64/aarch64-neon-builtins-base.cc b/gcc/config/aarch64/aarch64-neon-builtins-base.cc index 339df23504eb..6a9700ae8c3c 100644 --- a/gcc/config/aarch64/aarch64-neon-builtins-base.cc +++ b/gcc/config/aarch64/aarch64-neon-builtins-base.cc @@ -1623,6 +1623,11 @@ NEON_FUNCTION (vrndn, vrndnq, vrndns, gimple_ifn, (IFN_ROUNDEVEN)) NEON_FUNCTION (vrndp, vrndpq, gimple_ifn, (IFN_CEIL)) NEON_FUNCTION (vrndx, vrndxq, gimple_ifn, (IFN_RINT)) +// Shifts by an immediate +NEON_FUNCTION (vshl_n, vshlq_n, vshld_n, gimple_shift_imm, (LSHIFT_EXPR)) +NEON_FUNCTION (vshr_n, vshrq_n, vshrd_n, gimple_shift_imm, (RSHIFT_EXPR)) +NEON_FUNCTION (vsra_n, vsraq_n, vsrad_n, gimple_shift_imm, (RSHIFT_EXPR, PLUS_EXPR)) + // Bitwise operations NEON_FUNCTION (vand, vandq, gimple_expr, (BIT_AND_EXPR)) diff --git a/gcc/config/aarch64/aarch64-neon-builtins-base.def b/gcc/config/aarch64/aarch64-neon-builtins-base.def index 5b05d2f40348..d197f28b80aa 100644 --- a/gcc/config/aarch64/aarch64-neon-builtins-base.def +++ b/gcc/config/aarch64/aarch64-neon-builtins-base.def @@ -388,6 +388,19 @@ DEF_NEON_FUNCTION (vrbit, b_neon, ("D0,D0")) DEF_NEON_FUNCTION (vrbitq, b_neon, ("Q0,Q0")) #undef REQUIRED_EXTENSIONS +// Shifts by an immediate +#define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_SIMD) +DEF_NEON_FUNCTION (vshl_n, all_integer, ("D0,D0,ss32", shift<1>)) +DEF_NEON_FUNCTION (vshlq_n, all_integer, ("Q0,Q0,ss32", shift<1>)) +DEF_NEON_FUNCTION (vshr_n, all_integer, ("D0,D0,ss32", shift_right<1>)) +DEF_NEON_FUNCTION (vshrq_n, all_integer, ("Q0,Q0,ss32", shift_right<1>)) +DEF_NEON_FUNCTION (vsra_n, all_integer, ("D0,D0,D0,ss32", shift_right<2>)) +DEF_NEON_FUNCTION (vsraq_n, all_integer, ("Q0,Q0,Q0,ss32", shift_right<2>)) +DEF_NEON_FUNCTION (vshld_n, d_integer, ("s0,s0,ss32", shift<1>)) +DEF_NEON_FUNCTION (vshrd_n, d_integer, ("s0,s0,ss32", shift_right<1>)) +DEF_NEON_FUNCTION (vsrad_n, d_integer, ("s0,s0,s0,ss32", shift_right<2>)) +#undef REQUIRED_EXTENSIONS + // Bitwise operations (FEAT_SHA3) #define REQUIRED_EXTENSIONS nonstreaming_only (AARCH64_FL_SHA3) DEF_NEON_FUNCTION (vbcaxq, all_integer, ("Q0,Q0,Q0,Q0")) diff --git a/gcc/config/aarch64/aarch64-simd-builtins.def b/gcc/config/aarch64/aarch64-simd-builtins.def index aa7db1005191..b6bfb86efe65 100644 --- a/gcc/config/aarch64/aarch64-simd-builtins.def +++ b/gcc/config/aarch64/aarch64-simd-builtins.def @@ -381,7 +381,6 @@ BUILTIN_VSDQ_HSI (TERNOP_LANE, sqrdmulh_lane, 0, DEFAULT) BUILTIN_VSDQ_HSI (TERNOP_LANE, sqrdmulh_laneq, 0, DEFAULT) - BUILTIN_VSDQ_I_DI (BINOP, ashl, 3, DEFAULT) /* Implemented by aarch64_<sur>shl<mode>. */ BUILTIN_VSDQ_I_DI (BINOP, sshl, 0, DEFAULT) BUILTIN_VSDQ_I_DI (BINOP_UUS, ushl, 0, DEFAULT) @@ -420,17 +419,9 @@ BUILTIN_VQ_HSF (QUADOP_LANE_PAIR, fcmlaq_lane90, 0, FP) BUILTIN_VQ_HSF (QUADOP_LANE_PAIR, fcmlaq_lane180, 0, FP) BUILTIN_VQ_HSF (QUADOP_LANE_PAIR, fcmlaq_lane270, 0, FP) - - BUILTIN_VDQ_I (SHIFTIMM, ashr, 3, DEFAULT) - VAR1 (SHIFTIMM, ashr_simd, 0, DEFAULT, di) - BUILTIN_VDQ_I (USHIFTIMM, lshr, 3, DEFAULT) - VAR1 (USHIFTIMM, lshr_simd, 0, DEFAULT, di) /* Implemented by aarch64_<sur>shr_n<mode>. */ BUILTIN_VSDQ_I_DI (SHIFTIMM, srshr_n, 0, DEFAULT) BUILTIN_VSDQ_I_DI (USHIFTIMM, urshr_n, 0, DEFAULT) - /* Implemented by aarch64_<sur>sra_n<mode>. */ - BUILTIN_VSDQ_I_DI (SHIFTACC, ssra_n, 0, DEFAULT) - BUILTIN_VSDQ_I_DI (USHIFTACC, usra_n, 0, DEFAULT) BUILTIN_VSDQ_I_DI (SHIFTACC, srsra_n, 0, DEFAULT) BUILTIN_VSDQ_I_DI (USHIFTACC, ursra_n, 0, DEFAULT) /* Implemented by aarch64_<sur>shll_n<mode>. */ diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md index 08ea269c6939..c3db2dd05d51 100644 --- a/gcc/config/aarch64/aarch64-simd.md +++ b/gcc/config/aarch64/aarch64-simd.md @@ -1583,7 +1583,7 @@ } ) -(define_insn "aarch64_<sra_op>sra_n<mode>_insn" +(define_insn "*aarch64_<sra_op>sra_n<mode>" [(set (match_operand:VDQ_I 0 "register_operand" "=w") (plus:VDQ_I (SHIFTRT:VDQ_I @@ -1647,20 +1647,6 @@ [(set_attr "type" "neon_shift_acc<q>")] ) -(define_expand "aarch64_<sra_op>sra_n<mode>" - [(set (match_operand:VDQ_I 0 "register_operand") - (plus:VDQ_I - (SHIFTRT:VDQ_I - (match_operand:VDQ_I 2 "register_operand") - (match_operand:SI 3 "aarch64_simd_shift_imm_offset_<ve_mode>")) - (match_operand:VDQ_I 1 "register_operand")))] - "TARGET_SIMD" - { - operands[3] - = aarch64_simd_gen_const_vector_dup (<MODE>mode, UINTVAL (operands[3])); - } -) - (define_expand "@aarch64_<sra_op>rsra_n<mode>" [(match_operand:VSDQ_I_DI 0 "register_operand") (match_operand:VSDQ_I_DI 1 "register_operand") @@ -1855,23 +1841,6 @@ DONE; }) -;; DI vector shift -(define_expand "aarch64_ashr_simddi" - [(match_operand:DI 0 "register_operand") - (match_operand:DI 1 "register_operand") - (match_operand:SI 2 "aarch64_shift_imm64_di")] - "TARGET_SIMD" - { - /* An arithmetic shift right by 64 fills the result with copies of the sign - bit, just like asr by 63 - however the standard pattern does not handle - a shift by 64. */ - if (INTVAL (operands[2]) == 64) - operands[2] = GEN_INT (63); - emit_insn (gen_ashrdi3 (operands[0], operands[1], operands[2])); - DONE; - } -) - (define_expand "vlshr<mode>3" [(match_operand:VDQ_I 0 "register_operand") (match_operand:VDQ_I 1 "register_operand") @@ -1885,20 +1854,6 @@ DONE; }) -(define_expand "aarch64_lshr_simddi" - [(match_operand:DI 0 "register_operand") - (match_operand:DI 1 "register_operand") - (match_operand:SI 2 "aarch64_shift_imm64_di")] - "TARGET_SIMD" - { - if (INTVAL (operands[2]) == 64) - emit_move_insn (operands[0], const0_rtx); - else - emit_insn (gen_lshrdi3 (operands[0], operands[1], operands[2])); - DONE; - } -) - ;; For 64-bit modes we use ushl/r, as this does not require a SIMD zero. (define_insn "vec_shr_<mode><vczle><vczbe>" [(set (match_operand:VD 0 "register_operand" "=w") @@ -7434,20 +7389,6 @@ } ) -;; v(r)sra_n - -(define_insn "aarch64_<sur>sra_ndi" - [(set (match_operand:DI 0 "register_operand" "=w") - (unspec:DI [(match_operand:DI 1 "register_operand" "0") - (match_operand:DI 2 "register_operand" "w") - (match_operand:SI 3 - "aarch64_simd_shift_imm_offset_di" "i")] - VSRA))] - "TARGET_SIMD" - "<sur>sra\\t%d0, %d2, %3" - [(set_attr "type" "neon_shift_acc")] -) - ;; vs<lr>i_n (define_insn "@aarch64_<sur>s<lr>i_n<mode>" diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md index 9e7768e7fe1a..6916f6ec0aa9 100644 --- a/gcc/config/aarch64/aarch64.md +++ b/gcc/config/aarch64/aarch64.md @@ -3281,6 +3281,20 @@ [(set_attr "type" "alus_sreg")] ) +(define_insn "*aarch64_<sra_op>sra_ndi" + [(set (match_operand:DI 0 "register_operand") + (plus:DI + (SHIFTRT:DI + (match_operand:DI 1 "register_operand") + (match_operand:QI 2 "aarch64_simd_shift_imm_offset_di")) + (match_operand:DI 3 "register_operand")))] + "" + {@ [ cons: =0 , 1 , 2 , 3 ; attrs: type , arch ] + [ r , r , n , r ; alu_shift_imm_other , * ] add\t%x0, %x3, %x1, <shift> %2 + [ w , w , n , 0 ; neon_shift_acc , simd ] <sra_op>sra\t%d0, %d1, %2 + } +) + (define_insn "*add_<shift>_<mode>" [(set (match_operand:GPI 0 "register_operand" "=r") (plus:GPI (ASHIFT:GPI (match_operand:GPI 1 "register_operand" "r") diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h index b3a4080bb9ec..f263d164c9ed 100644 --- a/gcc/config/aarch64/arm_neon.h +++ b/gcc/config/aarch64/arm_neon.h @@ -12780,132 +12780,6 @@ vrsrad_n_u64 (uint64_t __a, uint64_t __b, const int __c) /* vshl */ -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_s8 (int8x8_t __a, const int __b) -{ - return (int8x8_t) __builtin_aarch64_ashlv8qi (__a, __b); -} - -__extension__ extern __inline int16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_s16 (int16x4_t __a, const int __b) -{ - return (int16x4_t) __builtin_aarch64_ashlv4hi (__a, __b); -} - -__extension__ extern __inline int32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_s32 (int32x2_t __a, const int __b) -{ - return (int32x2_t) __builtin_aarch64_ashlv2si (__a, __b); -} - -__extension__ extern __inline int64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_s64 (int64x1_t __a, const int __b) -{ - return (int64x1_t) {__builtin_aarch64_ashldi (__a[0], __b)}; -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_u8 (uint8x8_t __a, const int __b) -{ - return (uint8x8_t) __builtin_aarch64_ashlv8qi ((int8x8_t) __a, __b); -} - -__extension__ extern __inline uint16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_u16 (uint16x4_t __a, const int __b) -{ - return (uint16x4_t) __builtin_aarch64_ashlv4hi ((int16x4_t) __a, __b); -} - -__extension__ extern __inline uint32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_u32 (uint32x2_t __a, const int __b) -{ - return (uint32x2_t) __builtin_aarch64_ashlv2si ((int32x2_t) __a, __b); -} - -__extension__ extern __inline uint64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshl_n_u64 (uint64x1_t __a, const int __b) -{ - return (uint64x1_t) {__builtin_aarch64_ashldi ((int64_t) __a[0], __b)}; -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_s8 (int8x16_t __a, const int __b) -{ - return (int8x16_t) __builtin_aarch64_ashlv16qi (__a, __b); -} - -__extension__ extern __inline int16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_s16 (int16x8_t __a, const int __b) -{ - return (int16x8_t) __builtin_aarch64_ashlv8hi (__a, __b); -} - -__extension__ extern __inline int32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_s32 (int32x4_t __a, const int __b) -{ - return (int32x4_t) __builtin_aarch64_ashlv4si (__a, __b); -} - -__extension__ extern __inline int64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_s64 (int64x2_t __a, const int __b) -{ - return (int64x2_t) __builtin_aarch64_ashlv2di (__a, __b); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_u8 (uint8x16_t __a, const int __b) -{ - return (uint8x16_t) __builtin_aarch64_ashlv16qi ((int8x16_t) __a, __b); -} - -__extension__ extern __inline uint16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_u16 (uint16x8_t __a, const int __b) -{ - return (uint16x8_t) __builtin_aarch64_ashlv8hi ((int16x8_t) __a, __b); -} - -__extension__ extern __inline uint32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_u32 (uint32x4_t __a, const int __b) -{ - return (uint32x4_t) __builtin_aarch64_ashlv4si ((int32x4_t) __a, __b); -} - -__extension__ extern __inline uint64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshlq_n_u64 (uint64x2_t __a, const int __b) -{ - return (uint64x2_t) __builtin_aarch64_ashlv2di ((int64x2_t) __a, __b); -} - -__extension__ extern __inline int64_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshld_n_s64 (int64_t __a, const int __b) -{ - return __builtin_aarch64_ashldi (__a, __b); -} - -__extension__ extern __inline uint64_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshld_n_u64 (uint64_t __a, const int __b) -{ - return (uint64_t) __builtin_aarch64_ashldi (__a, __b); -} - __extension__ extern __inline int8x8_t __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) vshl_s8 (int8x8_t __a, int8x8_t __b) @@ -13116,134 +12990,6 @@ vshll_n_u32 (uint32x2_t __a, const int __b) return __builtin_aarch64_ushll_nv2si_uus (__a, __b); } -/* vshr */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_s8 (int8x8_t __a, const int __b) -{ - return (int8x8_t) __builtin_aarch64_ashrv8qi (__a, __b); -} - -__extension__ extern __inline int16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_s16 (int16x4_t __a, const int __b) -{ - return (int16x4_t) __builtin_aarch64_ashrv4hi (__a, __b); -} - -__extension__ extern __inline int32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_s32 (int32x2_t __a, const int __b) -{ - return (int32x2_t) __builtin_aarch64_ashrv2si (__a, __b); -} - -__extension__ extern __inline int64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_s64 (int64x1_t __a, const int __b) -{ - return (int64x1_t) {__builtin_aarch64_ashr_simddi (__a[0], __b)}; -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_u8 (uint8x8_t __a, const int __b) -{ - return __builtin_aarch64_lshrv8qi_uus (__a, __b); -} - -__extension__ extern __inline uint16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_u16 (uint16x4_t __a, const int __b) -{ - return __builtin_aarch64_lshrv4hi_uus (__a, __b); -} - -__extension__ extern __inline uint32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_u32 (uint32x2_t __a, const int __b) -{ - return __builtin_aarch64_lshrv2si_uus (__a, __b); -} - -__extension__ extern __inline uint64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshr_n_u64 (uint64x1_t __a, const int __b) -{ - return (uint64x1_t) {__builtin_aarch64_lshr_simddi_uus ( __a[0], __b)}; -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_s8 (int8x16_t __a, const int __b) -{ - return (int8x16_t) __builtin_aarch64_ashrv16qi (__a, __b); -} - -__extension__ extern __inline int16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_s16 (int16x8_t __a, const int __b) -{ - return (int16x8_t) __builtin_aarch64_ashrv8hi (__a, __b); -} - -__extension__ extern __inline int32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_s32 (int32x4_t __a, const int __b) -{ - return (int32x4_t) __builtin_aarch64_ashrv4si (__a, __b); -} - -__extension__ extern __inline int64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_s64 (int64x2_t __a, const int __b) -{ - return (int64x2_t) __builtin_aarch64_ashrv2di (__a, __b); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_u8 (uint8x16_t __a, const int __b) -{ - return __builtin_aarch64_lshrv16qi_uus (__a, __b); -} - -__extension__ extern __inline uint16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_u16 (uint16x8_t __a, const int __b) -{ - return __builtin_aarch64_lshrv8hi_uus (__a, __b); -} - -__extension__ extern __inline uint32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_u32 (uint32x4_t __a, const int __b) -{ - return __builtin_aarch64_lshrv4si_uus (__a, __b); -} - -__extension__ extern __inline uint64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrq_n_u64 (uint64x2_t __a, const int __b) -{ - return __builtin_aarch64_lshrv2di_uus (__a, __b); -} - -__extension__ extern __inline int64_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrd_n_s64 (int64_t __a, const int __b) -{ - return __builtin_aarch64_ashr_simddi (__a, __b); -} - -__extension__ extern __inline uint64_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vshrd_n_u64 (uint64_t __a, const int __b) -{ - return __builtin_aarch64_lshr_simddi_uus (__a, __b); -} - /* vsli */ __extension__ extern __inline int8x8_t @@ -13472,134 +13218,6 @@ vsqaddd_u64 (uint64_t __a, int64_t __b) return __builtin_aarch64_usqadddi_uus (__a, __b); } -/* vsra */ - -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_s8 (int8x8_t __a, int8x8_t __b, const int __c) -{ - return (int8x8_t) __builtin_aarch64_ssra_nv8qi (__a, __b, __c); -} - -__extension__ extern __inline int16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_s16 (int16x4_t __a, int16x4_t __b, const int __c) -{ - return (int16x4_t) __builtin_aarch64_ssra_nv4hi (__a, __b, __c); -} - -__extension__ extern __inline int32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_s32 (int32x2_t __a, int32x2_t __b, const int __c) -{ - return (int32x2_t) __builtin_aarch64_ssra_nv2si (__a, __b, __c); -} - -__extension__ extern __inline int64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_s64 (int64x1_t __a, int64x1_t __b, const int __c) -{ - return (int64x1_t) {__builtin_aarch64_ssra_ndi (__a[0], __b[0], __c)}; -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c) -{ - return __builtin_aarch64_usra_nv8qi_uuus (__a, __b, __c); -} - -__extension__ extern __inline uint16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c) -{ - return __builtin_aarch64_usra_nv4hi_uuus (__a, __b, __c); -} - -__extension__ extern __inline uint32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c) -{ - return __builtin_aarch64_usra_nv2si_uuus (__a, __b, __c); -} - -__extension__ extern __inline uint64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsra_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c) -{ - return (uint64x1_t) {__builtin_aarch64_usra_ndi_uuus (__a[0], __b[0], __c)}; -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c) -{ - return (int8x16_t) __builtin_aarch64_ssra_nv16qi (__a, __b, __c); -} - -__extension__ extern __inline int16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c) -{ - return (int16x8_t) __builtin_aarch64_ssra_nv8hi (__a, __b, __c); -} - -__extension__ extern __inline int32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c) -{ - return (int32x4_t) __builtin_aarch64_ssra_nv4si (__a, __b, __c); -} - -__extension__ extern __inline int64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c) -{ - return (int64x2_t) __builtin_aarch64_ssra_nv2di (__a, __b, __c); -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c) -{ - return __builtin_aarch64_usra_nv16qi_uuus (__a, __b, __c); -} - -__extension__ extern __inline uint16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c) -{ - return __builtin_aarch64_usra_nv8hi_uuus (__a, __b, __c); -} - -__extension__ extern __inline uint32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c) -{ - return __builtin_aarch64_usra_nv4si_uuus (__a, __b, __c); -} - -__extension__ extern __inline uint64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsraq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c) -{ - return __builtin_aarch64_usra_nv2di_uuus (__a, __b, __c); -} - -__extension__ extern __inline int64_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsrad_n_s64 (int64_t __a, int64_t __b, const int __c) -{ - return __builtin_aarch64_ssra_ndi (__a, __b, __c); -} - -__extension__ extern __inline uint64_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsrad_n_u64 (uint64_t __a, uint64_t __b, const int __c) -{ - return __builtin_aarch64_usra_ndi_uuus (__a, __b, __c); -} - /* vsri */ __extension__ extern __inline int8x8_t diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md index 979198bcaee0..d995299c68e4 100644 --- a/gcc/config/aarch64/iterators.md +++ b/gcc/config/aarch64/iterators.md @@ -980,8 +980,6 @@ UNSPEC_FMULX ; Used in aarch64-simd.md. UNSPEC_USQADD ; Used in aarch64-simd.md. UNSPEC_SUQADD ; Used in aarch64-simd.md. - UNSPEC_SSRA ; Used in aarch64-simd.md. - UNSPEC_USRA ; Used in aarch64-simd.md. UNSPEC_SRSHR ; Used in aarch64-simd.md. UNSPEC_URSHR ; Used in aarch64-simd.md. UNSPEC_SQSHLU ; Used in aarch64-simd.md. @@ -3802,7 +3800,6 @@ (define_int_iterator VQSHL [UNSPEC_SQSHL UNSPEC_UQSHL UNSPEC_SQRSHL UNSPEC_UQRSHL]) -(define_int_iterator VSRA [UNSPEC_SSRA UNSPEC_USRA]) (define_int_iterator VSLRI [UNSPEC_SSLI UNSPEC_USLI UNSPEC_SSRI UNSPEC_USRI]) @@ -4848,7 +4845,6 @@ (UNSPEC_USQADD "us") (UNSPEC_SUQADD "su") (UNSPEC_SSLI "s") (UNSPEC_USLI "u") (UNSPEC_SSRI "s") (UNSPEC_USRI "u") - (UNSPEC_USRA "u") (UNSPEC_SSRA "s") (UNSPEC_URSHR "ur") (UNSPEC_SRSHR "sr") (UNSPEC_SQSHLU "s") (UNSPEC_SQSHL "s") (UNSPEC_UQSHL "u") diff --git a/gcc/config/aarch64/predicates.md b/gcc/config/aarch64/predicates.md index 523974240f03..0f5ce8c96f75 100644 --- a/gcc/config/aarch64/predicates.md +++ b/gcc/config/aarch64/predicates.md @@ -275,10 +275,6 @@ (and (match_code "const_int") (match_test "UINTVAL (op) < 64"))) -(define_predicate "aarch64_shift_imm64_di" - (and (match_code "const_int") - (match_test "UINTVAL (op) <= 64"))) - (define_predicate "aarch64_reg_or_shift_imm_si" (ior (match_operand 0 "register_operand") (match_operand 0 "aarch64_shift_imm_si"))) diff --git a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vshl-opt-3.c b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vshl-opt-3.c index 86fcc7ae7f22..416b2e33b6df 100644 --- a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vshl-opt-3.c +++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vshl-opt-3.c @@ -8,4 +8,7 @@ int16x8_t foo (int16x8_t a) return vshrq_n_s16 (a, 16); } -/* { dg-final { scan-assembler-times {\tsshr\t.+, 16} 1 } } */ +/* An arithmetic shift right by the element width leaves every bit equal to + the sign bit, so it folds to a shift by one less, which in turn is just a + comparison against zero. */ +/* { dg-final { scan-assembler-times {\tcmlt\t.+, #0} 1 } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vshl_n.c b/gcc/testsuite/gcc.target/aarch64/neon/vshl_n.c new file mode 100644 index 000000000000..67fe93f07c22 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vshl_n.c @@ -0,0 +1,45 @@ +/* { dg-do compile } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +/* +** test_vshl_n_s8_3: +** shl v0\.8b, v0\.8b, 3 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshl_n_s8, int8x8_t, 3) + +/* +** test_vshlq_n_u16_15: +** shl v0\.8h, v0\.8h, 15 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshlq_n_u16, uint16x8_t, 15) + +/* +** test_vshlq_n_s64_1: +** add v0\.2d, v0\.2d, v0\.2d +** ret +*/ +TEST_UNIFORM_UNARY_N (vshlq_n_s64, int64x2_t, 1) + +/* +** test_vshl_n_s64_7: +** shl d0, d0, 7 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshl_n_s64, int64x1_t, 7) + +/* +** test_vshlq_n_u32_0: +** ret +*/ +TEST_UNIFORM_UNARY_N (vshlq_n_u32, uint32x4_t, 0) + +/* +** test_vshld_n_s64_3: +** lsl x0, x0, 3 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshld_n_s64, int64_t, 3) diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vshr_n.c b/gcc/testsuite/gcc.target/aarch64/neon/vshr_n.c new file mode 100644 index 000000000000..799827bc8eca --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vshr_n.c @@ -0,0 +1,60 @@ +/* { dg-do compile } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +/* +** test_vshr_n_s8_3: +** sshr v0\.8b, v0\.8b, 3 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshr_n_s8, int8x8_t, 3) + +/* +** test_vshrq_n_u16_15: +** ushr v0\.8h, v0\.8h, 15 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshrq_n_u16, uint16x8_t, 15) + +/* +** test_vshr_n_u64_7: +** ushr d0, d0, 7 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshr_n_u64, uint64x1_t, 7) + +/* +** test_vshr_n_u8_8: +** movi v0\.2s, 0 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshr_n_u8, uint8x8_t, 8) + +/* +** test_vshrq_n_u32_32: +** movi v0\.4s, 0 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshrq_n_u32, uint32x4_t, 32) + +/* +** test_vshr_n_s8_8: +** cmlt v0\.8b, v0\.8b, #0 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshr_n_s8, int8x8_t, 8) + +/* +** test_vshrq_n_s32_32: +** cmlt v0\.4s, v0\.4s, #0 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshrq_n_s32, int32x4_t, 32) + +/* +** test_vshrd_n_u64_64: +** mov w0, 0 +** ret +*/ +TEST_UNIFORM_UNARY_N (vshrd_n_u64, uint64_t, 64) diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vsra_n.c b/gcc/testsuite/gcc.target/aarch64/neon/vsra_n.c new file mode 100644 index 000000000000..1d191fe81e24 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vsra_n.c @@ -0,0 +1,52 @@ +/* { dg-do compile } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +/* +** test_vsra_n_s8_3: +** ssra v0\.8b, v1\.8b, 3 +** ret +*/ +TEST_UNIFORM_BINARY_N (vsra_n_s8, int8x8_t, 3) + +/* +** test_vsraq_n_u32_5: +** usra v0\.4s, v1\.4s, 5 +** ret +*/ +TEST_UNIFORM_BINARY_N (vsraq_n_u32, uint32x4_t, 5) + +/* +** test_vsraq_n_s64_63: +** ssra v0\.2d, v1\.2d, 63 +** ret +*/ +TEST_UNIFORM_BINARY_N (vsraq_n_s64, int64x2_t, 63) + +/* +** test_vsraq_n_u32_32: +** ret +*/ +TEST_UNIFORM_BINARY_N (vsraq_n_u32, uint32x4_t, 32) + +/* +** test_vsraq_n_s32_32: +** ssra v0\.4s, v1\.4s, 31 +** ret +*/ +TEST_UNIFORM_BINARY_N (vsraq_n_s32, int32x4_t, 32) + +/* +** test_vsrad_n_s64_3: +** add x0, x0, x1, asr 3 +** ret +*/ +TEST_UNIFORM_BINARY_N (vsrad_n_s64, int64_t, 3) + +/* +** test_vsrad_n_u64_5: +** add x0, x0, x1, lsr 5 +** ret +*/ +TEST_UNIFORM_BINARY_N (vsrad_n_u64, uint64_t, 5) diff --git a/gcc/testsuite/gcc.target/aarch64/scalar_intrinsics.c b/gcc/testsuite/gcc.target/aarch64/scalar_intrinsics.c index 094aaff7b1c1..0b15b47b6d73 100644 --- a/gcc/testsuite/gcc.target/aarch64/scalar_intrinsics.c +++ b/gcc/testsuite/gcc.target/aarch64/scalar_intrinsics.c @@ -9,7 +9,7 @@ : "w"(V1) \ : /* No clobbers */); -/* { dg-final { scan-assembler-times "\\tadd\\tx\[0-9\]+" 2 } } */ +/* { dg-final { scan-assembler-times "\\tadd\\tx\[0-9\]+" 4 } } */ uint64_t test_vaddd_u64 (uint64_t a, uint64_t b) @@ -873,7 +873,8 @@ test_vshrd_n_u64 (uint64_t a) return vshrd_n_u64 (a, 3); } -/* { dg-final { scan-assembler-times "\\tssra\\td\[0-9\]+" 1 } } */ +/* vsrad_n folds to a shifted add, which avoids moving the operands into FP + registers for the ssra. */ int64_t test_vsrad_n_s64 (int64_t a, int64_t b) @@ -881,8 +882,6 @@ test_vsrad_n_s64 (int64_t a, int64_t b) return vsrad_n_s64 (a, b, 2); } -/* { dg-final { scan-assembler-times "\\tusra\\td\[0-9\]+" 1 } } */ - uint64_t test_vsrad_n_u64 (uint64_t a, uint64_t b) { diff --git a/gcc/testsuite/gcc.target/aarch64/sshr64_1.c b/gcc/testsuite/gcc.target/aarch64/sshr64_1.c index ce18abbce3b5..761049d7a7b6 100644 --- a/gcc/testsuite/gcc.target/aarch64/sshr64_1.c +++ b/gcc/testsuite/gcc.target/aarch64/sshr64_1.c @@ -75,24 +75,6 @@ test_other_neg () return 0; } -int __attribute__ ((noinline)) -test_no_sshr0 () -{ - int64x1_t arg; - int64x1_t result; - int64_t got; - int64_t exp; - arg = vcreate_s64 (0x0000000080000000); - result = vshr_n_s64 (arg, 0); - got = vget_lane_s64 (result, 0); - exp = 0x0000000080000000; - /* Expect: "result" = 0x0000000080000000. */ - if (exp != got) - return 1; - return 0; -} - -/* { dg-final { scan-assembler-not "sshr\\td\[0-9\]+, d\[0-9\]+, 0" } } */ int main () { @@ -106,9 +88,6 @@ main () if (test_other_neg ()) abort (); - if (test_no_sshr0 ()) - abort (); - return 0; } diff --git a/gcc/testsuite/gcc.target/aarch64/ushr64_1.c b/gcc/testsuite/gcc.target/aarch64/ushr64_1.c index 6b079976da7b..86a021c9fbc2 100644 --- a/gcc/testsuite/gcc.target/aarch64/ushr64_1.c +++ b/gcc/testsuite/gcc.target/aarch64/ushr64_1.c @@ -18,12 +18,6 @@ test_vshr_n_u64_4 (uint64x1_t passed, uint64_t expected) return vget_lane_u64 (vshr_n_u64 (passed, 4), 0) != expected; } -int __attribute__ ((noinline)) -test_vshr_n_u64_0 (uint64x1_t passed, uint64_t expected) -{ - return vget_lane_u64 (vshr_n_u64 (passed, 0), 0) != expected; -} - int __attribute__ ((noinline)) test_vshrd_n_u64_64 (uint64_t passed, uint64_t expected) { @@ -36,15 +30,8 @@ test_vshrd_n_u64_4 (uint64_t passed, uint64_t expected) return vshrd_n_u64 (passed, 4) != expected; } -int __attribute__ ((noinline)) -test_vshrd_n_u64_0 (uint64_t passed, uint64_t expected) -{ - return vshrd_n_u64 (passed, 0) != expected; -} - /* { dg-final { (scan-assembler-times "ushr\\td\[0-9\]+, d\[0-9\]+, 4" 2) || \ (scan-assembler-times "lsr\\tx\[0-9\]+, x\[0-9\]+, 4" 2) } } */ -/* { dg-final { scan-assembler-not "ushr\\td\[0-9\]+, d\[0-9\]+, 0" } } */ int main (int argc, char *argv[]) @@ -60,9 +47,6 @@ main (int argc, char *argv[]) if (test_vshr_n_u64_4 (vcreate_u64 (0xffffffff80000000), 0x0ffffffff8000000)) abort (); - if (test_vshr_n_u64_0 (vcreate_u64 (0x0000000080000000), 0x0000000080000000)) - abort (); - /* Testing vshrd_n_u64. */ if (test_vshrd_n_u64_64 (0x0000000080000000, 0)) abort (); @@ -74,9 +58,6 @@ main (int argc, char *argv[]) if (test_vshrd_n_u64_4 (0xffffffff80000000, 0x0ffffffff8000000)) abort (); - if (test_vshrd_n_u64_0 (0x0000000080000000, 0x0000000080000000)) - abort (); - return 0; }
