https://gcc.gnu.org/g:0ea8000b86f01eec1e88ab278677e01a8e38c9de
commit r17-5890-g0ea8000b86f01eec1e88ab278677e01a8e38c9de Author: Karl Meakin <[email protected]> Date: Wed Jul 29 19:26:16 2026 +0000 aarch64: Port vsub intrinsics to pragma-based framework Port the `vsub` family of NEON intrinsics to the pragma-based framework, and add assembly tests for every variant. gcc/ChangeLog: * config/aarch64/aarch64-neon-builtins-base.cc (vsubh, vsubd, vsub, vsubq): New function bases. * config/aarch64/aarch64-neon-builtins-base.def (vsubd, vsub, vsubq, vsubh): New function declarations. * config/aarch64/arm_neon.h (vsub_s8, vsub_s16, vsub_s32, vsub_f32, vsub_f64, vsub_u8, vsub_u16, vsub_u32, vsub_s64, vsub_u64, vsubq_s8, vsubq_s16, vsubq_s32, vsubq_s64, vsubq_f32, vsubq_f64, vsubq_u8, vsubq_u16, vsubq_u32, vsubq_u64, vsub_f16, vsubq_f16): Delete function definitions. gcc/testsuite/ChangeLog: * gcc.target/aarch64/neon/vsub.c: New test. Diff: --- gcc/config/aarch64/aarch64-neon-builtins-base.cc | 1 + gcc/config/aarch64/aarch64-neon-builtins-base.def | 9 ++ gcc/config/aarch64/arm_neon.h | 154 ------------------- gcc/testsuite/gcc.target/aarch64/neon/vsub.c | 172 ++++++++++++++++++++++ 4 files changed, 182 insertions(+), 154 deletions(-) diff --git a/gcc/config/aarch64/aarch64-neon-builtins-base.cc b/gcc/config/aarch64/aarch64-neon-builtins-base.cc index ba5aa68748fd..74e68249f943 100644 --- a/gcc/config/aarch64/aarch64-neon-builtins-base.cc +++ b/gcc/config/aarch64/aarch64-neon-builtins-base.cc @@ -876,6 +876,7 @@ NEON_FUNCTION (vdupb_lane, vdupb_laneq, vduph_lane, vduph_laneq, // Lanewise arithmetic NEON_FUNCTION (vadd, vaddq, vaddd, gimple_arith, (PLUS_EXPR, PLUS_EXPR, BIT_XOR_EXPR)) +NEON_FUNCTION (vsub, vsubq, vsubd, gimple_arith, (MINUS_EXPR)) // Saturating arithmetic NEON_FUNCTION (vqaddb, vqaddh, vqadds, vqaddd, vqadd, vqaddq, gimple_ifn, (IFN_SAT_ADD)) diff --git a/gcc/config/aarch64/aarch64-neon-builtins-base.def b/gcc/config/aarch64/aarch64-neon-builtins-base.def index d7a6657e3f0c..2034416634ac 100644 --- a/gcc/config/aarch64/aarch64-neon-builtins-base.def +++ b/gcc/config/aarch64/aarch64-neon-builtins-base.def @@ -71,6 +71,11 @@ DEF_NEON_FUNCTION (vaddq, all_arith_no_fp16, ("Q0,Q0,Q0")) // Polynomial "addition" (really xor) DEF_NEON_FUNCTION (vadd, bhd_poly, ("D0,D0,D0")) DEF_NEON_FUNCTION (vaddq, bhdq_poly, ("Q0,Q0,Q0")) + +// Subtraction +DEF_NEON_FUNCTION (vsubd, d_integer, ("s0,s0,s0")) +DEF_NEON_FUNCTION (vsub, all_arith_no_fp16, ("D0,D0,D0")) +DEF_NEON_FUNCTION (vsubq, all_arith_no_fp16, ("Q0,Q0,Q0")) #undef REQUIRED_EXTENSIONS // Lanewise arithmetic (FP16) @@ -78,6 +83,10 @@ DEF_NEON_FUNCTION (vaddq, bhdq_poly, ("Q0,Q0,Q0")) // Addition DEF_NEON_FUNCTION (vadd, h_float, ("D0,D0,D0")) DEF_NEON_FUNCTION (vaddq, h_float, ("Q0,Q0,Q0")) + +// Subtraction +DEF_NEON_FUNCTION (vsub, h_float, ("D0,D0,D0")) +DEF_NEON_FUNCTION (vsubq, h_float, ("Q0,Q0,Q0")) #undef REQUIRED_EXTENSIONS // Saturating arithmetic diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h index 7b5926ca21b2..b0072059ccd9 100644 --- a/gcc/config/aarch64/arm_neon.h +++ b/gcc/config/aarch64/arm_neon.h @@ -595,146 +595,6 @@ vmulq_p8 (poly8x16_t __a, poly8x16_t __b) return __builtin_aarch64_pmulv16qi_ppp (__a, __b); } -__extension__ extern __inline int8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_s8 (int8x8_t __a, int8x8_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline int16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_s16 (int16x4_t __a, int16x4_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline int32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_s32 (int32x2_t __a, int32x2_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline float32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_f32 (float32x2_t __a, float32x2_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline float64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_f64 (float64x1_t __a, float64x1_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint8x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_u8 (uint8x8_t __a, uint8x8_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_u16 (uint16x4_t __a, uint16x4_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint32x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_u32 (uint32x2_t __a, uint32x2_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline int64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_s64 (int64x1_t __a, int64x1_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint64x1_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_u64 (uint64x1_t __a, uint64x1_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline int8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_s8 (int8x16_t __a, int8x16_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline int16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_s16 (int16x8_t __a, int16x8_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline int32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_s32 (int32x4_t __a, int32x4_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline int64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_s64 (int64x2_t __a, int64x2_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline float32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_f32 (float32x4_t __a, float32x4_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline float64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_f64 (float64x2_t __a, float64x2_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint8x16_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_u8 (uint8x16_t __a, uint8x16_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_u16 (uint16x8_t __a, uint16x8_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint32x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_u32 (uint32x4_t __a, uint32x4_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline uint64x2_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_u64 (uint64x2_t __a, uint64x2_t __b) -{ - return __a - __b; -} - __extension__ extern __inline int16x8_t __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) vsubl_s8 (int8x8_t __a, int8x8_t __b) @@ -18961,20 +18821,6 @@ vrsqrtsq_f16 (float16x8_t __a, float16x8_t __b) return __builtin_aarch64_rsqrtsv8hf (__a, __b); } -__extension__ extern __inline float16x4_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsub_f16 (float16x4_t __a, float16x4_t __b) -{ - return __a - __b; -} - -__extension__ extern __inline float16x8_t -__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) -vsubq_f16 (float16x8_t __a, float16x8_t __b) -{ - return __a - __b; -} - /* ARMv8.2-A FP16 three operands vector intrinsics. */ __extension__ extern __inline float16x4_t diff --git a/gcc/testsuite/gcc.target/aarch64/neon/vsub.c b/gcc/testsuite/gcc.target/aarch64/neon/vsub.c new file mode 100644 index 000000000000..2a0306f6aa72 --- /dev/null +++ b/gcc/testsuite/gcc.target/aarch64/neon/vsub.c @@ -0,0 +1,172 @@ +/* { dg-do compile } */ +/* { dg-final { check-function-bodies "**" "" } } */ + +#include "arm_neon_test.h" + +/* +** test_vsub_u8: +** sub v0\.8b, v0\.8b, v1\.8b +** ret +*/ +TEST_UNIFORM_BINARY (vsub_u8, uint8x8_t) + +/* +** test_vsub_s8: +** sub v0\.8b, v0\.8b, v1\.8b +** ret +*/ +TEST_UNIFORM_BINARY (vsub_s8, int8x8_t) + +/* +** test_vsub_u16: +** sub v0\.4h, v0\.4h, v1\.4h +** ret +*/ +TEST_UNIFORM_BINARY (vsub_u16, uint16x4_t) + +/* +** test_vsub_s16: +** sub v0\.4h, v0\.4h, v1\.4h +** ret +*/ +TEST_UNIFORM_BINARY (vsub_s16, int16x4_t) + +/* +** test_vsub_f16: +** fsub v0\.4h, v0\.4h, v1\.4h +** ret +*/ +TEST_UNIFORM_BINARY (vsub_f16, float16x4_t) + +/* +** test_vsub_u32: +** sub v0\.2s, v0\.2s, v1\.2s +** ret +*/ +TEST_UNIFORM_BINARY (vsub_u32, uint32x2_t) + +/* +** test_vsub_s32: +** sub v0\.2s, v0\.2s, v1\.2s +** ret +*/ +TEST_UNIFORM_BINARY (vsub_s32, int32x2_t) + +/* +** test_vsub_f32: +** fsub v0\.2s, v0\.2s, v1\.2s +** ret +*/ +TEST_UNIFORM_BINARY (vsub_f32, float32x2_t) + +/* +** test_vsub_u64: +** sub d0, d0, d1 +** ret +*/ +TEST_UNIFORM_BINARY (vsub_u64, uint64x1_t) + +/* +** test_vsub_s64: +** sub d0, d0, d1 +** ret +*/ +TEST_UNIFORM_BINARY (vsub_s64, int64x1_t) + +/* +** test_vsub_f64: +** fsub d0, d0, d1 +** ret +*/ +TEST_UNIFORM_BINARY (vsub_f64, float64x1_t) + +/* +** test_vsubq_u8: +** sub v0\.16b, v0\.16b, v1\.16b +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_u8, uint8x16_t) + +/* +** test_vsubq_s8: +** sub v0\.16b, v0\.16b, v1\.16b +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_s8, int8x16_t) + +/* +** test_vsubq_u16: +** sub v0\.8h, v0\.8h, v1\.8h +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_u16, uint16x8_t) + +/* +** test_vsubq_s16: +** sub v0\.8h, v0\.8h, v1\.8h +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_s16, int16x8_t) + +/* +** test_vsubq_f16: +** fsub v0\.8h, v0\.8h, v1\.8h +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_f16, float16x8_t) + +/* +** test_vsubq_u32: +** sub v0\.4s, v0\.4s, v1\.4s +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_u32, uint32x4_t) + +/* +** test_vsubq_s32: +** sub v0\.4s, v0\.4s, v1\.4s +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_s32, int32x4_t) + +/* +** test_vsubq_f32: +** fsub v0\.4s, v0\.4s, v1\.4s +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_f32, float32x4_t) + +/* +** test_vsubq_u64: +** sub v0\.2d, v0\.2d, v1\.2d +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_u64, uint64x2_t) + +/* +** test_vsubq_s64: +** sub v0\.2d, v0\.2d, v1\.2d +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_s64, int64x2_t) + +/* +** test_vsubq_f64: +** fsub v0\.2d, v0\.2d, v1\.2d +** ret +*/ +TEST_UNIFORM_BINARY (vsubq_f64, float64x2_t) + +/* +** test_vsubd_u64: +** sub x0, x0, x1 +** ret +*/ +TEST_UNIFORM_BINARY (vsubd_u64, uint64_t) + +/* +** test_vsubd_s64: +** sub x0, x0, x1 +** ret +*/ +TEST_UNIFORM_BINARY (vsubd_s64, int64_t)
