aarch64: Use RTL builtins for [su]mull_high_lane[q] intrinsics

Rewrite [su]mull_high_lane[q] Neon intrinsics to use RTL builtins
rather than inline assembly code, allowing for better scheduling and
optimization.

gcc/ChangeLog:

2021-02-03  Jonathan Wright  <jonathan.wright@arm.com>

	* config/aarch64/aarch64-simd-builtins.def: Add
	[su]mull_hi_lane[q] builtin generator macros.
	* config/aarch64/aarch64-simd.md
	(aarch64_<su>mull_hi_lane<mode>_insn): Define.
	(aarch64_<su>mull_hi_lane<mode>): Define.
	(aarch64_<su>mull_hi_laneq<mode>_insn): Define.
	(aarch64_<su>mull_hi_laneq<mode>): Define.
	* config/aarch64/arm_neon.h (vmull_high_lane_s16): Use RTL
	builtin instead of inline asm.
	(vmull_high_lane_s32): Likewise.
	(vmull_high_lane_u16): Likewise.
	(vmull_high_lane_u32): Likewise.
	(vmull_high_laneq_s16): Likewise.
	(vmull_high_laneq_s32): Likewise.
	(vmull_high_laneq_u16): Likewise.
	(vmull_high_laneq_u32): Liekwise.

3 files changed, 118 insertions, 96 deletions

diff --git a/gcc/config/aarch64/aarch64-simd-builtins.def b/gcc/config/aarch64/aarch64-simd-builtins.def
index 4dac8e0..29a7bbc 100644
--- a/gcc/config/aarch64/aarch64-simd-builtins.def
+++ b/gcc/config/aarch64/aarch64-simd-builtins.def
@@ -303,6 +303,11 @@
   BUILTIN_VQ_HSI (BINOP, smull_hi_n, 0, NONE)
   BUILTIN_VQ_HSI (BINOPU, umull_hi_n, 0, NONE)
 
+  BUILTIN_VQ_HSI (TERNOP_LANE, smull_hi_lane, 0, NONE)
+  BUILTIN_VQ_HSI (TERNOP_LANE, smull_hi_laneq, 0, NONE)
+  BUILTIN_VQ_HSI (TERNOPU_LANE, umull_hi_lane, 0, NONE)
+  BUILTIN_VQ_HSI (TERNOPU_LANE, umull_hi_laneq, 0, NONE)
+
   BUILTIN_VD_HSI (TERNOP_LANE, vec_smult_lane_, 0, NONE)
   BUILTIN_VD_HSI (QUADOP_LANE, vec_smlal_lane_, 0, NONE)
   BUILTIN_VD_HSI (TERNOP_LANE, vec_smult_laneq_, 0, NONE)
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 1d1ba37..60eeddc 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -2253,6 +2253,70 @@
   [(set_attr "type" "neon_mul_<Vetype>_scalar_long")]
 )
 
+(define_insn "aarch64_<su>mull_hi_lane<mode>_insn"
+  [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
+	(mult:<VWIDE>
+	  (ANY_EXTEND:<VWIDE> (vec_select:<VHALF>
+	    (match_operand:VQ_HSI 1 "register_operand" "w")
+	    (match_operand:VQ_HSI 2 "vect_par_cnst_hi_half" "")))
+	  (ANY_EXTEND:<VWIDE> (vec_duplicate:<VHALF>
+	    (vec_select:<VEL>
+	      (match_operand:<VCOND> 3 "register_operand" "<vwx>")
+	      (parallel [(match_operand:SI 4 "immediate_operand" "i")]))))))]
+  "TARGET_SIMD"
+  {
+    operands[4] = aarch64_endian_lane_rtx (<VCOND>mode, INTVAL (operands[4]));
+    return "<su>mull2\\t%0.<Vwtype>, %1.<Vtype>, %3.<Vetype>[%4]";
+  }
+  [(set_attr "type" "neon_mul_<Vetype>_scalar_long")]
+)
+
+(define_expand "aarch64_<su>mull_hi_lane<mode>"
+  [(match_operand:<VWIDE> 0 "register_operand")
+   (ANY_EXTEND:<VWIDE>(match_operand:VQ_HSI 1 "register_operand"))
+   (match_operand:<VCOND> 2 "register_operand")
+   (match_operand:SI 3 "immediate_operand")]
+  "TARGET_SIMD"
+{
+  rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
+  emit_insn (gen_aarch64_<su>mull_hi_lane<mode>_insn (operands[0],
+	     operands[1], p, operands[2], operands[3]));
+  DONE;
+}
+)
+
+(define_insn "aarch64_<su>mull_hi_laneq<mode>_insn"
+  [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
+	(mult:<VWIDE>
+	  (ANY_EXTEND:<VWIDE> (vec_select:<VHALF>
+	    (match_operand:VQ_HSI 1 "register_operand" "w")
+	    (match_operand:VQ_HSI 2 "vect_par_cnst_hi_half" "")))
+	  (ANY_EXTEND:<VWIDE> (vec_duplicate:<VHALF>
+	    (vec_select:<VEL>
+	      (match_operand:<VCONQ> 3 "register_operand" "<vwx>")
+	      (parallel [(match_operand:SI 4 "immediate_operand" "i")]))))))]
+  "TARGET_SIMD"
+  {
+    operands[4] = aarch64_endian_lane_rtx (<VCONQ>mode, INTVAL (operands[4]));
+    return "<su>mull2\\t%0.<Vwtype>, %1.<Vtype>, %3.<Vetype>[%4]";
+  }
+  [(set_attr "type" "neon_mul_<Vetype>_scalar_long")]
+)
+
+(define_expand "aarch64_<su>mull_hi_laneq<mode>"
+  [(match_operand:<VWIDE> 0 "register_operand")
+   (ANY_EXTEND:<VWIDE>(match_operand:VQ_HSI 1 "register_operand"))
+   (match_operand:<VCONQ> 2 "register_operand")
+   (match_operand:SI 3 "immediate_operand")]
+  "TARGET_SIMD"
+{
+  rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
+  emit_insn (gen_aarch64_<su>mull_hi_laneq<mode>_insn (operands[0],
+	     operands[1], p, operands[2], operands[3]));
+  DONE;
+}
+)
+
 (define_insn "aarch64_<su>mull_n<mode>"
   [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
         (mult:<VWIDE>
diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index fffd7b7..2d776ef 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -8167,109 +8167,62 @@ vshrn_n_u64 (uint64x2_t __a, const int __b)
 {
   return (uint32x2_t)__builtin_aarch64_shrnv2di ((int64x2_t)__a, __b);
 }
-#define vmull_high_lane_s16(a, b, c)                                    \
-  __extension__                                                         \
-    ({                                                                  \
-       int16x4_t b_ = (b);                                              \
-       int16x8_t a_ = (a);                                              \
-       int32x4_t result;                                                \
-       __asm__ ("smull2 %0.4s, %1.8h, %2.h[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "x"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
 
-#define vmull_high_lane_s32(a, b, c)                                    \
-  __extension__                                                         \
-    ({                                                                  \
-       int32x2_t b_ = (b);                                              \
-       int32x4_t a_ = (a);                                              \
-       int64x2_t result;                                                \
-       __asm__ ("smull2 %0.2d, %1.4s, %2.s[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "w"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_lane_s16 (int16x8_t __a, int16x4_t __v, const int __lane)
+{
+  return __builtin_aarch64_smull_hi_lanev8hi (__a, __v, __lane);
+}
 
-#define vmull_high_lane_u16(a, b, c)                                    \
-  __extension__                                                         \
-    ({                                                                  \
-       uint16x4_t b_ = (b);                                             \
-       uint16x8_t a_ = (a);                                             \
-       uint32x4_t result;                                               \
-       __asm__ ("umull2 %0.4s, %1.8h, %2.h[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "x"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_lane_s32 (int32x4_t __a, int32x2_t __v, const int __lane)
+{
+  return __builtin_aarch64_smull_hi_lanev4si (__a, __v, __lane);
+}
 
-#define vmull_high_lane_u32(a, b, c)                                    \
-  __extension__                                                         \
-    ({                                                                  \
-       uint32x2_t b_ = (b);                                             \
-       uint32x4_t a_ = (a);                                             \
-       uint64x2_t result;                                               \
-       __asm__ ("umull2 %0.2d, %1.4s, %2.s[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "w"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_lane_u16 (uint16x8_t __a, uint16x4_t __v, const int __lane)
+{
+  return __builtin_aarch64_umull_hi_lanev8hi_uuus (__a, __v, __lane);
+}
 
-#define vmull_high_laneq_s16(a, b, c)                                   \
-  __extension__                                                         \
-    ({                                                                  \
-       int16x8_t b_ = (b);                                              \
-       int16x8_t a_ = (a);                                              \
-       int32x4_t result;                                                \
-       __asm__ ("smull2 %0.4s, %1.8h, %2.h[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "x"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_lane_u32 (uint32x4_t __a, uint32x2_t __v, const int __lane)
+{
+  return __builtin_aarch64_umull_hi_lanev4si_uuus (__a, __v, __lane);
+}
 
-#define vmull_high_laneq_s32(a, b, c)                                   \
-  __extension__                                                         \
-    ({                                                                  \
-       int32x4_t b_ = (b);                                              \
-       int32x4_t a_ = (a);                                              \
-       int64x2_t result;                                                \
-       __asm__ ("smull2 %0.2d, %1.4s, %2.s[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "w"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_laneq_s16 (int16x8_t __a, int16x8_t __v, const int __lane)
+{
+  return __builtin_aarch64_smull_hi_laneqv8hi (__a, __v, __lane);
+}
 
-#define vmull_high_laneq_u16(a, b, c)                                   \
-  __extension__                                                         \
-    ({                                                                  \
-       uint16x8_t b_ = (b);                                             \
-       uint16x8_t a_ = (a);                                             \
-       uint32x4_t result;                                               \
-       __asm__ ("umull2 %0.4s, %1.8h, %2.h[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "x"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_laneq_s32 (int32x4_t __a, int32x4_t __v, const int __lane)
+{
+  return __builtin_aarch64_smull_hi_laneqv4si (__a, __v, __lane);
+}
 
-#define vmull_high_laneq_u32(a, b, c)                                   \
-  __extension__                                                         \
-    ({                                                                  \
-       uint32x4_t b_ = (b);                                             \
-       uint32x4_t a_ = (a);                                             \
-       uint64x2_t result;                                               \
-       __asm__ ("umull2 %0.2d, %1.4s, %2.s[%3]"                         \
-                : "=w"(result)                                          \
-                : "w"(a_), "w"(b_), "i"(c)                              \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_laneq_u16 (uint16x8_t __a, uint16x8_t __v, const int __lane)
+{
+  return __builtin_aarch64_umull_hi_laneqv8hi_uuus (__a, __v, __lane);
+}
+
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmull_high_laneq_u32 (uint32x4_t __a, uint32x4_t __v, const int __lane)
+{
+  return __builtin_aarch64_umull_hi_laneqv4si_uuus (__a, __v, __lane);
+}
 
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))