@@ -10316,6 +10316,15 @@ vld1_p64_x2 (const poly64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline poly64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_p64_x3 (const poly64_t * __a)
+{
+ union { poly64x1x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#pragma GCC pop_options
__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10381,6 +10390,42 @@ vld1_s64_x2 (const int64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline int8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_s8_x3 (const int8_t * __a)
+{
+ union { int8x8x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_s16_x3 (const int16_t * __a)
+{
+ union { int16x4x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_s32_x3 (const int32_t * __a)
+{
+ union { int32x2x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v2si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_s64_x3 (const int64_t * __a)
+{
+ union { int64x1x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline float16x4_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10417,6 +10462,26 @@ vld1_f32_x2 (const float32_t * __a)
return __rv.__i;
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline float16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_f16_x3 (const float16_t * __a)
+{
+ union { float16x4x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4hf (__a);
+ return __rv.__i;
+}
+#endif
+
+__extension__ extern __inline float32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_f32_x3 (const float32_t * __a)
+{
+ union { float32x2x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v2sf ((const __builtin_neon_sf *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline uint8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1_u8 (const uint8_t * __a)
@@ -10481,6 +10546,42 @@ vld1_u64_x2 (const uint64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline uint8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_u8_x3 (const uint8_t * __a)
+{
+ union { uint8x8x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_u16_x3 (const uint16_t * __a)
+{
+ union { uint16x4x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_u32_x3 (const uint32_t * __a)
+{
+ union { uint32x2x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v2si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_u64_x3 (const uint64_t * __a)
+{
+ union { uint64x1x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline poly8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1_p8 (const poly8_t * __a)
@@ -10513,6 +10614,24 @@ vld1_p16_x2 (const poly16_t * __a)
return __rv.__i;
}
+__extension__ extern __inline poly8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_p8_x3 (const poly8_t * __a)
+{
+ union { poly8x8x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline poly16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_p16_x3 (const poly16_t * __a)
+{
+ union { poly16x4x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
#pragma GCC push_options
#pragma GCC target ("fpu=crypto-neon-fp-armv8")
__extension__ extern __inline poly64x2_t
@@ -10536,7 +10655,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_p64_x3 (const poly64_t * __a)
{
union { poly64x2x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v2di ((const __builtin_neon_di *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v2di ((const __builtin_neon_di *) __a);
return __rv.__i;
}
@@ -10619,7 +10738,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_s8_x3 (const uint8_t * __a)
{
union { int8x16x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v16qi ((const __builtin_neon_qi *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v16qi ((const __builtin_neon_qi *) __a);
return __rv.__i;
}
@@ -10628,7 +10747,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_s16_x3 (const uint16_t * __a)
{
union { int16x8x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v8hi ((const __builtin_neon_hi *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v8hi ((const __builtin_neon_hi *) __a);
return __rv.__i;
}
@@ -10637,7 +10756,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_s32_x3 (const int32_t * __a)
{
union { int32x4x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v4si ((const __builtin_neon_si *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v4si ((const __builtin_neon_si *) __a);
return __rv.__i;
}
@@ -10646,7 +10765,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_s64_x3 (const int64_t * __a)
{
union { int64x2x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v2di ((const __builtin_neon_di *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v2di ((const __builtin_neon_di *) __a);
return __rv.__i;
}
@@ -10728,7 +10847,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_f16_x3 (const float16_t * __a)
{
union { float16x8x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v8hf (__a);
+ __rv.__o = __builtin_neon_vld1q_x3v8hf (__a);
return __rv.__i;
}
#endif
@@ -10738,7 +10857,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_f32_x3 (const float32_t * __a)
{
union { float32x4x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v4sf ((const __builtin_neon_sf *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v4sf ((const __builtin_neon_sf *) __a);
return __rv.__i;
}
@@ -10831,7 +10950,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u8_x3 (const uint8_t * __a)
{
union { uint8x16x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v16qi ((const __builtin_neon_qi *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v16qi ((const __builtin_neon_qi *) __a);
return __rv.__i;
}
@@ -10840,7 +10959,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u16_x3 (const uint16_t * __a)
{
union { uint16x8x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v8hi ((const __builtin_neon_hi *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v8hi ((const __builtin_neon_hi *) __a);
return __rv.__i;
}
@@ -10849,7 +10968,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u32_x3 (const uint32_t * __a)
{
union { uint32x4x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v4si ((const __builtin_neon_si *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v4si ((const __builtin_neon_si *) __a);
return __rv.__i;
}
@@ -10858,7 +10977,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u64_x3 (const uint64_t * __a)
{
union { uint64x2x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v2di ((const __builtin_neon_di *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v2di ((const __builtin_neon_di *) __a);
return __rv.__i;
}
@@ -10935,7 +11054,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_p8_x3 (const poly8_t * __a)
{
union { poly8x16x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v16qi ((const __builtin_neon_qi *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v16qi ((const __builtin_neon_qi *) __a);
return __rv.__i;
}
@@ -10944,7 +11063,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_p16_x3 (const poly16_t * __a)
{
union { poly16x8x3_t __i; __builtin_neon_ci __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v8hi ((const __builtin_neon_hi *) __a);
+ __rv.__o = __builtin_neon_vld1q_x3v8hi ((const __builtin_neon_hi *) __a);
return __rv.__i;
}
@@ -20944,6 +21063,15 @@ vld1_bf16_x2 (const bfloat16_t * __ptr)
return __rv.__i;
}
+__extension__ extern __inline bfloat16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1_bf16_x3 (const bfloat16_t * __ptr)
+{
+ union { bfloat16x4x3_t __i; __builtin_neon_ei __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4bf ((const __builtin_neon_bf *) __ptr);
+ return __rv.__i;
+}
+
__extension__ extern __inline bfloat16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_bf16 (const bfloat16_t * __ptr)
@@ -20965,7 +21093,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_bf16_x3 (const bfloat16_t * __ptr)
{
union { bfloat16x8x3_t __i; __builtin_neon_oi __o; } __rv;
- __rv.__o = __builtin_neon_vld1_x3v8bf ((const __builtin_neon_bf *) __ptr);
+ __rv.__o = __builtin_neon_vld1q_x3v8bf ((const __builtin_neon_bf *) __ptr);
return __rv.__i;
}
@@ -303,7 +303,8 @@ VAR13 (LOAD1, vld1,
v4bf, v8bf)
VAR7 (LOAD1, vld1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
VAR7 (LOAD1, vld1q_x2, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
-VAR7 (LOAD1, vld1_x3, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
+VAR7 (LOAD1, vld1_x3, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
+VAR7 (LOAD1, vld1q_x3, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
VAR7 (LOAD1, vld1_x4, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
VAR12 (LOAD1LANE, vld1_lane,
v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di, v4bf, v8bf)
@@ -4967,7 +4967,17 @@ if (BYTES_BIG_ENDIAN)
[(set_attr "type" "neon_load1_2reg<q>")]
)
-(define_expand "neon_vld1_x3<mode>"
+(define_insn "neon_vld1_x3<mode>"
+ [(set (match_operand:EI 0 "s_register_operand" "=w")
+ (unspec:EI [(match_operand:EI 1 "neon_struct_operand" "Um")
+ (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VLD1))]
+ "TARGET_NEON"
+ "vld1.<V_sz_elem>\t%h0, %A1"
+ [(set_attr "type" "neon_load1_3reg<q>")]
+)
+
+(define_expand "neon_vld1q_x3<mode>"
[(match_operand:CI 0 "s_register_operand")
(match_operand:CI 1 "neon_struct_operand")
(unspec:VQXBF [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
@@ -60,7 +60,62 @@ poly16x4x2_t test_vld1_p16_x2 (poly16_t * a)
return vld1_p16_x2 (a);
}
-/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
-/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
-/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
-/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
\ No newline at end of file
+uint8x8x3_t test_vld1_u8_x3 (uint8_t * a)
+{
+ return vld1_u8_x3 (a);
+}
+
+uint16x4x3_t test_vld1_u16_x3 (uint16_t * a)
+{
+ return vld1_u16_x3 (a);
+}
+
+uint32x2x3_t test_vld1_u32_x3 (uint32_t * a)
+{
+ return vld1_u32_x3 (a);
+}
+
+uint64x1x3_t test_vld1_u64_x3 (uint64_t * a)
+{
+ return vld1_u64_x3 (a);
+}
+
+int8x8x3_t test_vld1_s8_x3 (int8_t * a)
+{
+ return vld1_s8_x3 (a);
+}
+
+int16x4x3_t test_vld1_s16_x3 (int16_t * a)
+{
+ return vld1_s16_x3 (a);
+}
+
+int32x2x3_t test_vld1_s32_x3 (int32_t * a)
+{
+ return vld1_s32_x3 (a);
+}
+
+int64x1x3_t test_vld1_s64_x3 (int64_t * a)
+{
+ return vld1_s64_x3 (a);
+}
+
+float32x2x3_t test_vld1_f32_x3 (float32_t * a)
+{
+ return vld1_f32_x3 (a);
+}
+
+poly8x8x3_t test_vld1_p8_x3 (poly8_t * a)
+{
+ return vld1_p8_x3 (a);
+}
+
+poly16x4x3_t test_vld1_p16_x3 (poly16_t * a)
+{
+ return vld1_p16_x3 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
\ No newline at end of file
@@ -10,4 +10,9 @@ bfloat16x4x2_t test_vld1_bf16_x2 (bfloat16_t * a)
return vld1_bf16_x2 (a);
}
-/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
\ No newline at end of file
+bfloat16x4x3_t test_vld1_bf16_x3 (bfloat16_t * a)
+{
+ return vld1_bf16_x3 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
\ No newline at end of file
@@ -10,4 +10,9 @@ float16x4x2_t test_vld1_f16_x2 (float16_t * a)
return vld1_f16_x2 (a);
}
-/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+float16x4x3_t test_vld1_f16_x3 (float16_t * a)
+{
+ return vld1_f16_x3 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
@@ -10,4 +10,9 @@ poly64x1x2_t test_vld1_p64_x2 (poly64_t * a)
return vld1_p64_x2 (a);
}
-/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
+poly64x1x3_t test_vld1_p64_x3 (poly64_t * a)
+{
+ return vld1_p64_x3 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
From: Ezra Sitorus <ezra.sitorus@arm.com> This patch is part of a series of patches implementing the _xN variants of the vld1 intrinsic for the arm port. This patch adds the _x3 variants of the vld1 intrinsic. The previous vld1_x3 has been updated to vld1q_x3 to take into account that it works with 4-word-length types. vld1_x3 is now only for 2-word-length types. 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 (vld1_u8_x3, vld1_u16_x3, vld1_u32_x3, vld1_u64_x3): New. (vld1_s8_x3, vld1_s16_x3, vld1_s32_x3, vld1_s64_x3): New. (vld1_f16_x3, vld1_f32_x3): New. (vld1_p8_x3, vld1_p16_x3, vld1_p64_x3): New. (vld1_bf16_x3): New. (vld1q_types_x3): Updated to use vld1q_x3 from arm_neon_builtins.def * config/arm/arm_neon_builtins.def (vld1_x3): Updated entries. (vld1q_x3): New entries, but comes from the old vld1_x2 * config/arm/neon.md (neon_vld1q_x3<mode>): Updated from neon_vld1_x3<mode>. gcc/testsuite/ChangeLog: * gcc.target/arm/simd/vld1_base_xN_1.c: Updated. * gcc.target/arm/simd/vld1_bf16_xN_1.c: Updated. * gcc.target/arm/simd/vld1_fp16_xN_1.c: Updated. * gcc.target/arm/simd/vld1_p64_xN_1.c: Updated. --- gcc/config/arm/arm_neon.h | 156 ++++++++++++++++-- gcc/config/arm/arm_neon_builtins.def | 3 +- gcc/config/arm/neon.md | 12 +- .../gcc.target/arm/simd/vld1_base_xN_1.c | 63 ++++++- .../gcc.target/arm/simd/vld1_bf16_xN_1.c | 7 +- .../gcc.target/arm/simd/vld1_fp16_xN_1.c | 7 +- .../gcc.target/arm/simd/vld1_p64_xN_1.c | 7 +- 7 files changed, 232 insertions(+), 23 deletions(-)