arm: vld1q_types_x3 ACLE intrinsics

This patch is part of a series of patches implementing the _xN
variants of the vld1q intrinsic for the arm port. This patch adds the
_x3 variants of the vld1q 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
	(vld1q_u8_x3, vld1q_u16_x3, vld1q_u32_x3, vld1q_u64_x3): New.
	(vld1q_s8_x3, vld1q_s16_x3, vld1q_s32_x3, vld1q_s64_x3): New.
	(vld1q_f16_x3, vld1q_f32_x3): New.
	(vld1q_p8_x3, vld1q_p16_x3, vld1q_p64_x3): New.
	(vld1q_bf16_x3): New.
	* config/arm/arm_neon_builtins.def (vld1_x3): New entries.
	* config/arm/neon.md (vld1_x3<mode>): New.

gcc/testsuite/ChangeLog:
	* gcc.target/arm/simd/vld1q_base_xN_1.c: Add new tests.
	* gcc.target/arm/simd/vld1q_bf16_xN_1.c: Add new tests.
	* gcc.target/arm/simd/vld1q_fp16_xN_1.c: Add new tests.
	* gcc.target/arm/simd/vld1q_p64_xN_1.c: Add new tests.

7 files changed, 236 insertions, 3 deletions

diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index 3eb41c6..557873a 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -10412,6 +10412,15 @@ vld1q_p64_x2 (const poly64_t * __a)
   return __rv.__i;
 }
 
+__extension__ extern __inline poly64x2x3_t
+__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);
+  return __rv.__i;
+}
+
 #pragma GCC pop_options
 __extension__ extern __inline int8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10477,6 +10486,42 @@ vld1q_s64_x2 (const int64_t * __a)
   return __rv.__i;
 }
 
+__extension__ extern __inline int8x16x3_t
+__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);
+  return __rv.__i;
+}
+
+__extension__ extern __inline int16x8x3_t
+__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);
+  return __rv.__i;
+}
+
+__extension__ extern __inline int32x4x3_t
+__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);
+  return __rv.__i;
+}
+
+__extension__ extern __inline int64x2x3_t
+__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);
+  return __rv.__i;
+}
+
 #if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
 __extension__ extern __inline float16x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10513,6 +10558,26 @@ vld1q_f32_x2 (const float32_t * __a)
   return __rv.__i;
 }
 
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline float16x8x3_t
+__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);
+  return __rv.__i;
+}
+#endif
+
+__extension__ extern __inline float32x4x3_t
+__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);
+  return __rv.__i;
+}
+
 __extension__ extern __inline uint8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_u8 (const uint8_t * __a)
@@ -10577,6 +10642,42 @@ vld1q_u64_x2 (const uint64_t * __a)
   return __rv.__i;
 }
 
+__extension__ extern __inline uint8x16x3_t
+__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);
+  return __rv.__i;
+}
+
+__extension__ extern __inline uint16x8x3_t
+__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);
+  return __rv.__i;
+}
+
+__extension__ extern __inline uint32x4x3_t
+__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);
+  return __rv.__i;
+}
+
+__extension__ extern __inline uint64x2x3_t
+__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);
+  return __rv.__i;
+}
+
 __extension__ extern __inline poly8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_p8 (const poly8_t * __a)
@@ -10609,6 +10710,24 @@ vld1q_p16_x2 (const poly16_t * __a)
   return __rv.__i;
 }
 
+__extension__ extern __inline poly8x16x3_t
+__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);
+  return __rv.__i;
+}
+
+__extension__ extern __inline poly16x8x3_t
+__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);
+  return __rv.__i;
+}
+
 __extension__ extern __inline int8x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_s8 (const int8_t * __a, int8x8_t __b, const int __c)
@@ -19910,6 +20029,15 @@ vld1q_bf16_x2 (const bfloat16_t * __ptr)
   return __rv.__i;
 }
 
+__extension__ extern __inline bfloat16x8x3_t
+__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);
+  return __rv.__i;
+}
+
 __extension__ extern __inline bfloat16x4x2_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_bf16 (bfloat16_t const * __ptr)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index 6a8f0cb..a363bf1 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -302,6 +302,7 @@ VAR13 (LOAD1, vld1,
         v8qi, v4hi, v4hf, v2si, v2sf, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
         v4bf, v8bf)
 VAR7 (LOAD1, vld1_x2, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
+VAR7 (LOAD1, vld1_x3, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
 VAR12 (LOAD1LANE, vld1_lane,
 	v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di, v4bf, v8bf)
 VAR10 (LOAD1, vld1_dup,
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index 55049ea..b37d95f 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -4967,6 +4967,33 @@ if (BYTES_BIG_ENDIAN)
   [(set_attr "type" "neon_load1_2reg<q>")]
 )
 
+(define_insn "neon_vld1_x3<mode>"
+  [(set (match_operand:CI 0 "s_register_operand" "=w")
+        (unspec:CI [(match_operand:EI 1 "neon_struct_operand" "Um")
+                    (unspec:VQXBF [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+                   UNSPEC_VLD3A))]
+  "TARGET_NEON"
+{
+  int regno = REGNO (operands[0]);
+  rtx ops[4];
+  ops[0] = gen_rtx_REG (DImode, regno);
+  ops[1] = gen_rtx_REG (DImode, regno + 2);
+  ops[2] = gen_rtx_REG (DImode, regno + 4);
+  ops[3] = operands[1];
+
+  output_asm_insn ("vld1.<V_sz_elem>\t{%P0, %P1, %P2}, %A3", ops);
+
+  ops[0] = gen_rtx_REG (DImode, regno + 6);
+  ops[1] = gen_rtx_REG (DImode, regno + 8);
+  ops[2] = gen_rtx_REG (DImode, regno + 10);
+  ops[3] = operands[1];
+
+  output_asm_insn ("vld1.<V_sz_elem>\t{%P0, %P1, %P2}, %A3", ops);
+  return "";
+}
+  [(set_attr "type" "neon_load1_3reg<q>")]
+)
+
 ;; The lane numbers in the RTL are in GCC lane order, having been flipped
 ;; in arm_expand_neon_args. The lane numbers are restored to architectural
 ;; lane order here.
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
index 1d31777..bfad282 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
@@ -60,8 +60,69 @@ poly16x8x2_t test_vld1q_p16_x2 (poly16_t * a)
     return vld1q_p16_x2 (a);
 }
 
+uint8x16x3_t test_vld1q_u8_x3 (uint8_t * a)
+{
+    return vld1q_u8_x3 (a);
+}
+
+uint16x8x3_t test_vld1q_u16_x3 (uint16_t * a)
+{
+    return vld1q_u16_x3 (a);
+}
+
+uint32x4x3_t test_vld1q_u32_x3 (uint32_t * a)
+{
+    return vld1q_u32_x3 (a);
+}
+
+uint64x2x3_t test_vld1q_u64_x3 (uint64_t * a)
+{
+    return vld1q_u64_x3 (a);
+}
+
+int8x16x3_t test_vld1q_s8_x3 (int8_t * a)
+{
+    return vld1q_s8_x3 (a);
+}
+
+int16x8x3_t test_vld1q_s16_x3 (int16_t * a)
+{
+    return vld1q_s16_x3 (a);
+}
+
+int32x4x3_t test_vld1q_s32_x3 (int32_t * a)
+{
+    return vld1q_s32_x3 (a);
+}
+
+int64x2x3_t test_vld1q_s64_x3 (int64_t * a)
+{
+    return vld1q_s64_x3 (a);
+}
+
+float32x4x3_t test_vld1q_f32_x3 (float32_t * a)
+{
+    return vld1q_f32_x3 (a);
+}
+
+poly8x16x3_t test_vld1q_p8_x3 (poly8_t * a)
+{
+    return vld1q_p8_x3 (a);
+}
+
+poly16x8x3_t test_vld1q_p16_x3 (poly16_t * a)
+{
+    return vld1q_p16_x3 (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 {vld1.8\t\{d[0-9]+, 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} 3 } }  */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, 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} 3 } }  */
-/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } }  */
+/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+, 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} 2 } }  */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } }  */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
index 5f6fc98..4138fe9 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
@@ -10,4 +10,10 @@ bfloat16x8x2_t test_vld1q_bf16_x2 (bfloat16_t * a)
     return vld1q_bf16_x2 (a);
 }
 
+bfloat16x8x3_t test_vld1q_bf16_x3 (bfloat16_t * a)
+{
+    return vld1q_bf16_x3 (a);
+}
+
 /* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } }  */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } }  */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
index aecf491..01640d7 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
@@ -10,5 +10,10 @@ float16x8x2_t test_vld1q_f16_x2 (float16_t * a)
     return vld1q_f16_x2 (a);
 }
 
-/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } }  */
+float16x8x3_t test_vld1q_f16_x3 (float16_t * a)
+{
+    return vld1q_f16_x3 (a);
+}
 
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } }  */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } }  */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
index 04ceb5e..ae2ab36 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
@@ -10,5 +10,10 @@ poly64x2x2_t test_vld1q_p64_x2 (poly64_t * a)
     return vld1q_p64_x2 (a);
 }
 
-/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } }  */
+poly64x2x3_t test_vld1q_p64_x3 (poly64_t * a)
+{
+    return vld1q_p64_x3 (a);
+}
 
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } }  */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } }  */