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

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

gcc/ChangeLog:

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

	* config/aarch64/aarch64-simd-builtins.def: Add
	[su]mlal_hi_lane[q] builtin generator macros.
	* config/aarch64/aarch64-simd.md
	(aarch64_<su>mlal_hi_lane<mode>_insn): Define.
	(aarch64_<su>mlal_hi_lane<mode>): Define.
	(aarch64_<su>mlal_hi_laneq<mode>_insn): Define.
	(aarch64_<su>mlal_hi_laneq<mode>): Define.
	* config/aarch64/arm_neon.h (vmlal_high_lane_s16): Use RTL
	builtin instead of inline asm.
	(vmlal_high_lane_s32): Likewise.
	(vmlal_high_lane_u16): Likewise.
	(vmlal_high_lane_u32): Likewise.
	(vmlal_high_laneq_s16): Likewise.
	(vmlal_high_laneq_s32): Likewise.
	(vmlal_high_laneq_u16): Likewise.
	(vmlal_high_laneq_u32): Likewise.

3 files changed, 131 insertions, 104 deletions

diff --git a/gcc/config/aarch64/aarch64-simd-builtins.def b/gcc/config/aarch64/aarch64-simd-builtins.def
index 319cd64..55cc2a4 100644
--- a/gcc/config/aarch64/aarch64-simd-builtins.def
+++ b/gcc/config/aarch64/aarch64-simd-builtins.def
@@ -314,6 +314,11 @@
   BUILTIN_VD_HSI (QUADOPU_LANE, vec_umlsl_lane_, 0, NONE)
   BUILTIN_VD_HSI (QUADOPU_LANE, vec_umlsl_laneq_, 0, NONE)
 
+  BUILTIN_VQ_HSI (QUADOP_LANE, smlal_hi_lane, 0, NONE)
+  BUILTIN_VQ_HSI (QUADOP_LANE, smlal_hi_laneq, 0, NONE)
+  BUILTIN_VQ_HSI (QUADOPU_LANE, umlal_hi_lane, 0, NONE)
+  BUILTIN_VQ_HSI (QUADOPU_LANE, umlal_hi_laneq, 0, NONE)
+
   BUILTIN_VSD_HSI (BINOP, sqdmull, 0, NONE)
   BUILTIN_VSD_HSI (TERNOP_LANE, sqdmull_lane, 0, NONE)
   BUILTIN_VSD_HSI (TERNOP_LANE, sqdmull_laneq, 0, NONE)
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index fd506bc..60ecd13 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -2287,6 +2287,76 @@
   [(set_attr "type" "neon_mla_<Vetype>_scalar_long")]
 )
 
+(define_insn "aarch64_<su>mlal_hi_lane<mode>_insn"
+  [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
+	(plus:<VWIDE>
+	  (mult:<VWIDE>
+	    (ANY_EXTEND:<VWIDE> (vec_select:<VHALF>
+	      (match_operand:VQ_HSI 2 "register_operand" "w")
+	      (match_operand:VQ_HSI 3 "vect_par_cnst_hi_half" "")))
+	    (ANY_EXTEND:<VWIDE> (vec_duplicate:<VHALF>
+	      (vec_select:<VEL>
+		(match_operand:<VCOND> 4 "register_operand" "<vwx>")
+		(parallel [(match_operand:SI 5 "immediate_operand" "i")])))))
+	  (match_operand:<VWIDE> 1 "register_operand" "0")))]
+  "TARGET_SIMD"
+  {
+    operands[5] = aarch64_endian_lane_rtx (<VCOND>mode, INTVAL (operands[5]));
+    return "<su>mlal2\\t%0.<Vwtype>, %2.<Vtype>, %4.<Vetype>[%5]";
+  }
+  [(set_attr "type" "neon_mla_<Vetype>_scalar_long")]
+)
+
+(define_expand "aarch64_<su>mlal_hi_lane<mode>"
+  [(match_operand:<VWIDE> 0 "register_operand")
+   (match_operand:<VWIDE> 1 "register_operand")
+   (ANY_EXTEND:<VWIDE>(match_operand:VQ_HSI 2 "register_operand"))
+   (match_operand:<VCOND> 3 "register_operand")
+   (match_operand:SI 4 "immediate_operand")]
+  "TARGET_SIMD"
+{
+  rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
+  emit_insn (gen_aarch64_<su>mlal_hi_lane<mode>_insn (operands[0],
+	     operands[1], operands[2], p, operands[3], operands[4]));
+  DONE;
+}
+)
+
+(define_insn "aarch64_<su>mlal_hi_laneq<mode>_insn"
+  [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
+	(plus:<VWIDE>
+	  (mult:<VWIDE>
+	    (ANY_EXTEND:<VWIDE> (vec_select:<VHALF>
+	      (match_operand:VQ_HSI 2 "register_operand" "w")
+	      (match_operand:VQ_HSI 3 "vect_par_cnst_hi_half" "")))
+	    (ANY_EXTEND:<VWIDE> (vec_duplicate:<VHALF>
+	      (vec_select:<VEL>
+		(match_operand:<VCONQ> 4 "register_operand" "<vwx>")
+		(parallel [(match_operand:SI 5 "immediate_operand" "i")])))))
+	  (match_operand:<VWIDE> 1 "register_operand" "0")))]
+  "TARGET_SIMD"
+  {
+    operands[5] = aarch64_endian_lane_rtx (<VCONQ>mode, INTVAL (operands[5]));
+    return "<su>mlal2\\t%0.<Vwtype>, %2.<Vtype>, %4.<Vetype>[%5]";
+  }
+  [(set_attr "type" "neon_mla_<Vetype>_scalar_long")]
+)
+
+(define_expand "aarch64_<su>mlal_hi_laneq<mode>"
+  [(match_operand:<VWIDE> 0 "register_operand")
+   (match_operand:<VWIDE> 1 "register_operand")
+   (ANY_EXTEND:<VWIDE>(match_operand:VQ_HSI 2 "register_operand"))
+   (match_operand:<VCONQ> 3 "register_operand")
+   (match_operand:SI 4 "immediate_operand")]
+  "TARGET_SIMD"
+{
+  rtx p = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
+  emit_insn (gen_aarch64_<su>mlal_hi_laneq<mode>_insn (operands[0],
+	     operands[1], operands[2], p, operands[3], operands[4]));
+  DONE;
+}
+)
+
 (define_insn "aarch64_vec_<su>mlsl_lane<Qlane>"
   [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
    (minus:<VWIDE>
diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index 11e6ca5..bfe6b2b 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -7152,117 +7152,69 @@ vmla_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
                                                  (int32x2_t) __c);
 }
 
-#define vmlal_high_lane_s16(a, b, c, d)                                 \
-  __extension__                                                         \
-    ({                                                                  \
-       int16x4_t c_ = (c);                                              \
-       int16x8_t b_ = (b);                                              \
-       int32x4_t a_ = (a);                                              \
-       int32x4_t result;                                                \
-       __asm__ ("smlal2 %0.4s, %2.8h, %3.h[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "x"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_lane_s16(int32x4_t __a, int16x8_t __b, int16x4_t __v,
+		    const int __lane)
+{
+  return __builtin_aarch64_smlal_hi_lanev8hi (__a, __b, __v, __lane);
+}
 
-#define vmlal_high_lane_s32(a, b, c, d)                                 \
-  __extension__                                                         \
-    ({                                                                  \
-       int32x2_t c_ = (c);                                              \
-       int32x4_t b_ = (b);                                              \
-       int64x2_t a_ = (a);                                              \
-       int64x2_t result;                                                \
-       __asm__ ("smlal2 %0.2d, %2.4s, %3.s[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "w"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_lane_s32(int64x2_t __a, int32x4_t __b, int32x2_t __v,
+		    const int __lane)
+{
+  return __builtin_aarch64_smlal_hi_lanev4si (__a, __b, __v, __lane);
+}
 
-#define vmlal_high_lane_u16(a, b, c, d)                                 \
-  __extension__                                                         \
-    ({                                                                  \
-       uint16x4_t c_ = (c);                                             \
-       uint16x8_t b_ = (b);                                             \
-       uint32x4_t a_ = (a);                                             \
-       uint32x4_t result;                                               \
-       __asm__ ("umlal2 %0.4s, %2.8h, %3.h[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "x"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_lane_u16(uint32x4_t __a, uint16x8_t __b, uint16x4_t __v,
+		    const int __lane)
+{
+  return __builtin_aarch64_umlal_hi_lanev8hi_uuuus (__a, __b, __v, __lane);
+}
 
-#define vmlal_high_lane_u32(a, b, c, d)                                 \
-  __extension__                                                         \
-    ({                                                                  \
-       uint32x2_t c_ = (c);                                             \
-       uint32x4_t b_ = (b);                                             \
-       uint64x2_t a_ = (a);                                             \
-       uint64x2_t result;                                               \
-       __asm__ ("umlal2 %0.2d, %2.4s, %3.s[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "w"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_lane_u32(uint64x2_t __a, uint32x4_t __b, uint32x2_t __v,
+		    const int __lane)
+{
+  return __builtin_aarch64_umlal_hi_lanev4si_uuuus (__a, __b, __v, __lane);
+}
 
-#define vmlal_high_laneq_s16(a, b, c, d)                                \
-  __extension__                                                         \
-    ({                                                                  \
-       int16x8_t c_ = (c);                                              \
-       int16x8_t b_ = (b);                                              \
-       int32x4_t a_ = (a);                                              \
-       int32x4_t result;                                                \
-       __asm__ ("smlal2 %0.4s, %2.8h, %3.h[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "x"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_laneq_s16(int32x4_t __a, int16x8_t __b, int16x8_t __v,
+		     const int __lane)
+{
+  return __builtin_aarch64_smlal_hi_laneqv8hi (__a, __b, __v, __lane);
+}
 
-#define vmlal_high_laneq_s32(a, b, c, d)                                \
-  __extension__                                                         \
-    ({                                                                  \
-       int32x4_t c_ = (c);                                              \
-       int32x4_t b_ = (b);                                              \
-       int64x2_t a_ = (a);                                              \
-       int64x2_t result;                                                \
-       __asm__ ("smlal2 %0.2d, %2.4s, %3.s[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "w"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_laneq_s32(int64x2_t __a, int32x4_t __b, int32x4_t __v,
+		     const int __lane)
+{
+  return __builtin_aarch64_smlal_hi_laneqv4si (__a, __b, __v, __lane);
+}
 
-#define vmlal_high_laneq_u16(a, b, c, d)                                \
-  __extension__                                                         \
-    ({                                                                  \
-       uint16x8_t c_ = (c);                                             \
-       uint16x8_t b_ = (b);                                             \
-       uint32x4_t a_ = (a);                                             \
-       uint32x4_t result;                                               \
-       __asm__ ("umlal2 %0.4s, %2.8h, %3.h[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "x"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_laneq_u16(uint32x4_t __a, uint16x8_t __b, uint16x8_t __v,
+		     const int __lane)
+{
+  return __builtin_aarch64_umlal_hi_laneqv8hi_uuuus (__a, __b, __v, __lane);
+}
 
-#define vmlal_high_laneq_u32(a, b, c, d)                                \
-  __extension__                                                         \
-    ({                                                                  \
-       uint32x4_t c_ = (c);                                             \
-       uint32x4_t b_ = (b);                                             \
-       uint64x2_t a_ = (a);                                             \
-       uint64x2_t result;                                               \
-       __asm__ ("umlal2 %0.2d, %2.4s, %3.s[%4]"                         \
-                : "=w"(result)                                          \
-                : "0"(a_), "w"(b_), "w"(c_), "i"(d)                     \
-                : /* No clobbers */);                                   \
-       result;                                                          \
-     })
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vmlal_high_laneq_u32(uint64x2_t __a, uint32x4_t __b, uint32x4_t __v,
+		     const int __lane)
+{
+  return __builtin_aarch64_umlal_hi_laneqv4si_uuuus (__a, __b, __v, __lane);
+}
 
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))