From: Ezra Sitorus Date: Tue, 2 Jan 2024 09:23:39 +0000 (+0000) Subject: arm: vst1_types_x4 ACLE intrinsics X-Git-Tag: basepoints/gcc-15~2933 X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=221912d34eeb42ada3fd5b4afdf72d0b4c85ee03;p=thirdparty%2Fgcc.git arm: vst1_types_x4 ACLE intrinsics This patch is part of a series of patches implementing the _xN variants of the vst1 intrinsic for the arm port. This patch adds the _x4 variants of the vst1 intrinsic. ACLE documents: https://developer.arm.com/documentation/ihi0053/latest/ ISA documents: https://developer.arm.com/documentation/ddi0487/latest/ gcc/ChangeLog: * config/arm/arm_neon.h (vst1_u8_x4, vst1_u16_x4, vst1_u32_x4, vst1_u64_x4): New. (vst1_s8_x4, vst1_s16_x4, vst1_s32_x4, vst1_s64_x4): New. (vst1_f16_x4, vst1_f32_x4): New. (vst1_p8_x4, vst1_p16_x4, vst1_p64_x4): New. (vst1_bf16_x4): New. * config/arm/arm_neon_builtins.def (vst1_x4): New entries. * config/arm/neon.md (vst1_x4): New. gcc/testsuite/ChangeLog: * gcc.target/arm/simd/vst1_base_xN_1.c: Updated. * gcc.target/arm/simd/vst1_bf16_xN_1.c: Updated. * gcc.target/arm/simd/vst1_fp16_xN_1.c: Updated. * gcc.target/arm/simd/vst1_p64_xN_1.c: Updated. --- diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h index d6527031bfa9..81c3579b7ef7 100644 --- a/gcc/config/arm/arm_neon.h +++ b/gcc/config/arm/arm_neon.h @@ -11258,6 +11258,14 @@ vst1_p64_x3 (poly64_t * __a, poly64x1x3_t __b) __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o); } +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_p64_x4 (poly64_t * __a, poly64x1x4_t __b) +{ + union { poly64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o); +} + #pragma GCC pop_options __extension__ extern __inline void __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) @@ -11351,6 +11359,38 @@ vst1_s64_x3 (int64_t * __a, int64x1x3_t __b) __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o); } +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_s8_x4 (int8_t * __a, int8x8x4_t __b) +{ + union { int8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o); +} + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_s16_x4 (int16_t * __a, int16x4x4_t __b) +{ + union { int16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o); +} + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_s32_x4 (int32_t * __a, int32x2x4_t __b) +{ + union { int32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v2si ((__builtin_neon_si *) __a, __bu.__o); +} + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_s64_x4 (int64_t * __a, int64x1x4_t __b) +{ + union { int64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4di ((__builtin_neon_di *) __a, __bu.__o); +} + #if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE) __extension__ extern __inline void __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) @@ -11403,6 +11443,24 @@ vst1_f32_x3 (float32_t * __a, float32x2x3_t __b) __builtin_neon_vst1_x3v2sf ((__builtin_neon_sf *) __a, __bu.__o); } +#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE) +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_f16_x4 (float16_t * __a, float16x4x4_t __b) +{ + union { float16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v4hf (__a, __bu.__o); +} +#endif + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_f32_x4 (float32_t * __a, float32x2x4_t __b) +{ + union { float32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v2sf ((__builtin_neon_sf *) __a, __bu.__o); +} + __extension__ extern __inline void __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) vst1_u8 (uint8_t * __a, uint8x8_t __b) @@ -11495,6 +11553,38 @@ vst1_u64_x3 (uint64_t * __a, uint64x1x3_t __b) __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o); } +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_u8_x4 (uint8_t * __a, uint8x8x4_t __b) +{ + union { uint8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o); +} + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_u16_x4 (uint16_t * __a, uint16x4x4_t __b) +{ + union { uint16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o); +} + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_u32_x4 (uint32_t * __a, uint32x2x4_t __b) +{ + union { uint32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v2si ((__builtin_neon_si *) __a, __bu.__o); +} + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_u64_x4 (uint64_t * __a, uint64x1x4_t __b) +{ + union { uint64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4di ((__builtin_neon_di *) __a, __bu.__o); +} + __extension__ extern __inline void __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) vst1_p8 (poly8_t * __a, poly8x8_t __b) @@ -11541,6 +11631,22 @@ vst1_p16_x3 (poly16_t * __a, poly16x4x3_t __b) __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o); } +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_p8_x4 (poly8_t * __a, poly8x8x4_t __b) +{ + union { poly8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o); +} + +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_p16_x4 (poly16_t * __a, poly16x4x4_t __b) +{ + union { poly16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o); +} + #pragma GCC push_options #pragma GCC target ("fpu=crypto-neon-fp-armv8") __extension__ extern __inline void @@ -20298,6 +20404,14 @@ vst1_bf16_x3 (bfloat16_t * __a, bfloat16x4x3_t __b) __builtin_neon_vst1_x3v4bf ((__builtin_neon_bf *) __a, __bu.__o); } +__extension__ extern __inline void +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) +vst1_bf16_x4 (bfloat16_t * __a, bfloat16x4x4_t __b) +{ + union { bfloat16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b }; + __builtin_neon_vst1_x4v4bf ((__builtin_neon_bf *) __a, __bu.__o); +} + __extension__ extern __inline void __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b) diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def index d4c684571f6a..327bdcacddf7 100644 --- a/gcc/config/arm/arm_neon_builtins.def +++ b/gcc/config/arm/arm_neon_builtins.def @@ -313,6 +313,7 @@ VAR14 (STORE1, vst1, v4bf, v8bf) VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf) VAR7 (STORE1, vst1_x3, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf) +VAR7 (STORE1, vst1_x4, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf) VAR14 (STORE1LANE, vst1_lane, v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si, v4sf, v2di, v4bf, v8bf) VAR13 (LOAD1, vld2, diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md index 4b6c63462b77..a5fb959308ef 100644 --- a/gcc/config/arm/neon.md +++ b/gcc/config/arm/neon.md @@ -5241,6 +5241,16 @@ if (BYTES_BIG_ENDIAN) [(set_attr "type" "neon_store1_3reg")] ) +(define_insn "neon_vst1_x4" + [(set (match_operand:OI 0 "neon_struct_operand" "=Um") + (unspec:OI [(match_operand:OI 1 "s_register_operand" "w") + (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)] + UNSPEC_VST1))] + "TARGET_NEON" + "vst1.\t%h1, %A0" + [(set_attr "type" "neon_store1_4reg")] +) + ;; see comment on neon_vld1_lane for reason why the lane numbers are reversed ;; here on big endian targets. (define_insn "neon_vst1_lane" diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c index 5f820a6a496e..04ca6583552f 100644 --- a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c @@ -115,8 +115,62 @@ void test_vst1_p16_x3 (poly16_t * ptr, poly16x4x3_t val) vst1_p16_x3 (ptr, val); } +void test_vst1_u8_x4 (uint8_t * ptr, uint8x8x4_t val) +{ + vst1_u8_x4 (ptr, val); +} + +void test_vst1_u16_x4 (uint16_t * ptr, uint16x4x4_t val) +{ + vst1_u16_x4 (ptr, val); +} + +void test_vst1_u32_x4 (uint32_t * ptr, uint32x2x4_t val) +{ + vst1_u32_x4 (ptr, val); +} + +void test_vst1_u64_x4 (uint64_t * ptr, uint64x1x4_t val) +{ + vst1_u64_x4 (ptr, val); +} + +void test_vst1_s8_x4 (int8_t * ptr, int8x8x4_t val) +{ + vst1_s8_x4 (ptr, val); +} + +void test_vst1_s16_x4 (int16_t * ptr, int16x4x4_t val) +{ + vst1_s16_x4 (ptr, val); +} + +void test_vst1_s32_x4 (int32_t * ptr, int32x2x4_t val) +{ + vst1_s32_x4 (ptr, val); +} + +void test_vst1_s64_x4 (int64_t * ptr, int64x1x4_t val) +{ + vst1_s64_x4 (ptr, val); +} + +void test_vst1_f32_x4 (float32_t * ptr, float32x2x4_t val) +{ + vst1_f32_x4 (ptr, val); +} + +void test_vst1_p8_x4 (poly8_t * ptr, poly8x8x4_t val) +{ + vst1_p8_x4 (ptr, val); +} + +void test_vst1_p16_x4 (poly16_t * ptr, poly16x4x4_t val) +{ + vst1_p16_x4 (ptr, val); +} -/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */ -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */ -/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */ -/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */ +/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */ +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */ +/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */ +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 6 } } */ diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c index a3a00ead4682..d919c7d060dc 100644 --- a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c @@ -15,4 +15,8 @@ void test_vst1_bf16_x3 (bfloat16_t * ptr, bfloat16x4x3_t val) vst1_bf16_x3 (ptr, val); } -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */ +void test_vst1_bf16_x4 (bfloat16_t * ptr, bfloat16x4x4_t val) +{ + vst1_bf16_x4 (ptr, val); +} +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */ diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c index 0a6863e24c6c..3d1d1eb7ad14 100644 --- a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c @@ -15,4 +15,9 @@ void test_vst1_f16_x3 (float16_t * ptr, float16x4x3_t val) vst1_f16_x3 (ptr, val); } -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */ +void test_vst1_f16_x4 (float16_t * ptr, float16x4x4_t val) +{ + vst1_f16_x4 (ptr, val); +} + +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */ diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c index 5dbd6049bc9f..62912143481a 100644 --- a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c @@ -15,4 +15,9 @@ void test_vst1_p64_x3 (poly64_t * ptr, poly64x1x3_t val) vst1_p64_x3 (ptr, val); } -/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */ \ No newline at end of file +void test_vst1_p64_x4 (poly64_t * ptr, poly64x1x4_t val) +{ + vst1_p64_x4 (ptr, val); +} + +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 3 } } */ \ No newline at end of file