aarch64: Use RTL builtins for integer mla_n intrinsics

Rewrite integer mla_n Neon intrinsics to use RTL builtins rather than
inline assembly code, allowing for better scheduling and
optimization.

gcc/ChangeLog:

2021-01-15  Jonathan Wright  <jonathan.wright@arm.com>

	* config/aarch64/aarch64-simd-builtins.def: Add mla_n builtin
	generator macro.
	* config/aarch64/aarch64-simd.md (*aarch64_mla_elt_merge<mode>):
	Rename to...
	(aarch64_mla_n<mode>): This.
	* config/aarch64/arm_neon.h (vmla_n_s16): Use RTL builtin
	instead of asm.
	(vmla_n_s32): Likewise.
	(vmla_n_u16): Likewise.
	(vmla_n_u32): Likewise.
	(vmlaq_n_s16): Likewise.
	(vmlaq_n_s32): Likewise.
	(vmlaq_n_u16): Likewise.
	(vmlaq_n_u32): Likewise.

3 files changed, 26 insertions, 55 deletions

diff --git a/gcc/config/aarch64/aarch64-simd-builtins.def b/gcc/config/aarch64/aarch64-simd-builtins.def
index a233156..ef83d5e 100644
--- a/gcc/config/aarch64/aarch64-simd-builtins.def
+++ b/gcc/config/aarch64/aarch64-simd-builtins.def
@@ -180,6 +180,8 @@
 
   /* Implemented by aarch64_mla<mode>.  */
   BUILTIN_VDQ_BHSI (TERNOP, mla, 0, NONE)
+  /* Implemented by aarch64_mla_n<mode>.  */
+  BUILTIN_VDQHS (TERNOP, mla_n, 0, NONE)
 
   /* Implemented by aarch64_<su>mlsl<mode>.  */
   BUILTIN_VD_BHSI (TERNOP, smlsl, 0, NONE)
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index be2a5a86..693a618 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -1384,15 +1384,16 @@
   [(set_attr "type" "neon_mla_<Vetype>_scalar<q>")]
 )
 
-(define_insn "*aarch64_mla_elt_merge<mode>"
-  [(set (match_operand:VDQHS 0 "register_operand" "=w")
+(define_insn "aarch64_mla_n<mode>"
+ [(set (match_operand:VDQHS 0 "register_operand" "=w")
 	(plus:VDQHS
-	  (mult:VDQHS (vec_duplicate:VDQHS
-		  (match_operand:<VEL> 1 "register_operand" "<h_con>"))
-		(match_operand:VDQHS 2 "register_operand" "w"))
-	  (match_operand:VDQHS 3 "register_operand" "0")))]
+	  (mult:VDQHS
+	    (vec_duplicate:VDQHS
+	      (match_operand:<VEL> 3 "register_operand" "<h_con>"))
+	    (match_operand:VDQHS 2 "register_operand" "w"))
+	  (match_operand:VDQHS 1 "register_operand" "0")))]
  "TARGET_SIMD"
- "mla\t%0.<Vtype>, %2.<Vtype>, %1.<Vetype>[0]"
+ "mla\t%0.<Vtype>, %2.<Vtype>, %3.<Vetype>[0]"
   [(set_attr "type" "neon_mla_<Vetype>_scalar<q>")]
 )
 
diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index 15fb345..ef86562 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -7246,48 +7246,32 @@ __extension__ extern __inline int16x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_s16 (int16x4_t __a, int16x4_t __b, int16_t __c)
 {
-  int16x4_t __result;
-  __asm__ ("mla %0.4h,%2.4h,%3.h[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "x"(__c)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_mla_nv4hi (__a, __b, __c);
 }
 
 __extension__ extern __inline int32x2_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_s32 (int32x2_t __a, int32x2_t __b, int32_t __c)
 {
-  int32x2_t __result;
-  __asm__ ("mla %0.2s,%2.2s,%3.s[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "w"(__c)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_mla_nv2si (__a, __b, __c);
 }
 
 __extension__ extern __inline uint16x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_u16 (uint16x4_t __a, uint16x4_t __b, uint16_t __c)
 {
-  uint16x4_t __result;
-  __asm__ ("mla %0.4h,%2.4h,%3.h[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "x"(__c)
-           : /* No clobbers */);
-  return __result;
+  return (uint16x4_t) __builtin_aarch64_mla_nv4hi ((int16x4_t) __a,
+                                                   (int16x4_t) __b,
+                                                   (int16_t) __c);
 }
 
 __extension__ extern __inline uint32x2_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_u32 (uint32x2_t __a, uint32x2_t __b, uint32_t __c)
 {
-  uint32x2_t __result;
-  __asm__ ("mla %0.2s,%2.2s,%3.s[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "w"(__c)
-           : /* No clobbers */);
-  return __result;
+  return (uint32x2_t) __builtin_aarch64_mla_nv2si ((int32x2_t) __a,
+                                                   (int32x2_t) __b,
+                                                   (int32_t) __c);
 }
 
 __extension__ extern __inline int8x8_t
@@ -7763,48 +7747,32 @@ __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_s16 (int16x8_t __a, int16x8_t __b, int16_t __c)
 {
-  int16x8_t __result;
-  __asm__ ("mla %0.8h,%2.8h,%3.h[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "x"(__c)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_mla_nv8hi (__a, __b, __c);
 }
 
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_s32 (int32x4_t __a, int32x4_t __b, int32_t __c)
 {
-  int32x4_t __result;
-  __asm__ ("mla %0.4s,%2.4s,%3.s[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "w"(__c)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_mla_nv4si (__a, __b, __c);
 }
 
 __extension__ extern __inline uint16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_u16 (uint16x8_t __a, uint16x8_t __b, uint16_t __c)
 {
-  uint16x8_t __result;
-  __asm__ ("mla %0.8h,%2.8h,%3.h[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "x"(__c)
-           : /* No clobbers */);
-  return __result;
+  return (uint16x8_t) __builtin_aarch64_mla_nv8hi ((int16x8_t) __a,
+                                                   (int16x8_t) __b,
+                                                   (int16_t) __c);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_u32 (uint32x4_t __a, uint32x4_t __b, uint32_t __c)
 {
-  uint32x4_t __result;
-  __asm__ ("mla %0.4s,%2.4s,%3.s[0]"
-           : "=w"(__result)
-           : "0"(__a), "w"(__b), "w"(__c)
-           : /* No clobbers */);
-  return __result;
+  return (uint32x4_t) __builtin_aarch64_mla_nv4si ((int32x4_t) __a,
+                                                   (int32x4_t) __b,
+                                                   (int32_t) __c);
 }
 
 __extension__ extern __inline int8x16_t