[AArch64, 4/4] Reimplement vmvn* intrinscis, remove inline assembly

gcc/
	* config/aarch64/arm_neon.h (vmvn_s8): Reimplement using C operator.
	Remove inline assembly.
	(vmvn_s16): Likewise.
	(vmvn_s32): Likewise.
	(vmvn_u8): Likewise.
	(vmvn_u16): Likewise.
	(vmvn_u32): Likewise.
	(vmvnq_s8): Likewise.
	(vmvnq_s16): Likewise.
	(vmvnq_s32): Likewise.
	(vmvnq_u8): Likewise.
	(vmvnq_u16): Likewise.
	(vmvnq_u32): Likewise.
	(vmvn_p8): Likewise.
	(vmvnq_p16): Likewise.

From-SVN: r236334

2 files changed, 104 insertions, 155 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 2030c85..b1d1bc0 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,23 @@
 2016-05-17  Jiong Wang  <jiong.wang@arm.com>
 
+	* config/aarch64/arm_neon.h (vmvn_s8): Reimplement using C operator.
+	Remove inline assembly.
+	(vmvn_s16): Likewise.
+	(vmvn_s32): Likewise.
+	(vmvn_u8): Likewise.
+	(vmvn_u16): Likewise.
+	(vmvn_u32): Likewise.
+	(vmvnq_s8): Likewise.
+	(vmvnq_s16): Likewise.
+	(vmvnq_s32): Likewise.
+	(vmvnq_u8): Likewise.
+	(vmvnq_u16): Likewise.
+	(vmvnq_u32): Likewise.
+	(vmvn_p8): Likewise.
+	(vmvnq_p16): Likewise.
+
+2016-05-17  Jiong Wang  <jiong.wang@arm.com>
+
 	* config/aarch64/aarch64-simd.md (vmul_n_f32): Remove inline assembly.
 	Use builtin.
 	(vmul_n_s16): Likewise.
diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index 84931ae..e563e3d 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -8388,161 +8388,6 @@ vmull_u32 (uint32x2_t a, uint32x2_t b)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
-vmvn_p8 (poly8x8_t a)
-{
-  poly8x8_t result;
-  __asm__ ("mvn %0.8b,%1.8b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
-vmvn_s8 (int8x8_t a)
-{
-  int8x8_t result;
-  __asm__ ("mvn %0.8b,%1.8b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
-vmvn_s16 (int16x4_t a)
-{
-  int16x4_t result;
-  __asm__ ("mvn %0.8b,%1.8b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
-vmvn_s32 (int32x2_t a)
-{
-  int32x2_t result;
-  __asm__ ("mvn %0.8b,%1.8b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
-vmvn_u8 (uint8x8_t a)
-{
-  uint8x8_t result;
-  __asm__ ("mvn %0.8b,%1.8b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
-vmvn_u16 (uint16x4_t a)
-{
-  uint16x4_t result;
-  __asm__ ("mvn %0.8b,%1.8b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
-vmvn_u32 (uint32x2_t a)
-{
-  uint32x2_t result;
-  __asm__ ("mvn %0.8b,%1.8b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
-vmvnq_p8 (poly8x16_t a)
-{
-  poly8x16_t result;
-  __asm__ ("mvn %0.16b,%1.16b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
-vmvnq_s8 (int8x16_t a)
-{
-  int8x16_t result;
-  __asm__ ("mvn %0.16b,%1.16b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
-vmvnq_s16 (int16x8_t a)
-{
-  int16x8_t result;
-  __asm__ ("mvn %0.16b,%1.16b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
-vmvnq_s32 (int32x4_t a)
-{
-  int32x4_t result;
-  __asm__ ("mvn %0.16b,%1.16b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
-vmvnq_u8 (uint8x16_t a)
-{
-  uint8x16_t result;
-  __asm__ ("mvn %0.16b,%1.16b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
-vmvnq_u16 (uint16x8_t a)
-{
-  uint16x8_t result;
-  __asm__ ("mvn %0.16b,%1.16b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
-vmvnq_u32 (uint32x4_t a)
-{
-  uint32x4_t result;
-  __asm__ ("mvn %0.16b,%1.16b"
-           : "=w"(result)
-           : "w"(a)
-           : /* No clobbers */);
-  return result;
-}
-
-
 __extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
 vpadal_s8 (int16x4_t a, int8x8_t b)
 {
@@ -18871,6 +18716,92 @@ vmulq_n_u32 (uint32x4_t __a, uint32_t __b)
   return __a * __b;
 }
 
+/* vmvn  */
+
+__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+vmvn_p8 (poly8x8_t __a)
+{
+  return (poly8x8_t) ~((int8x8_t) __a);
+}
+
+__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+vmvn_s8 (int8x8_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+vmvn_s16 (int16x4_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+vmvn_s32 (int32x2_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+vmvn_u8 (uint8x8_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+vmvn_u16 (uint16x4_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+vmvn_u32 (uint32x2_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+vmvnq_p8 (poly8x16_t __a)
+{
+  return (poly8x16_t) ~((int8x16_t) __a);
+}
+
+__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+vmvnq_s8 (int8x16_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+vmvnq_s16 (int16x8_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+vmvnq_s32 (int32x4_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+vmvnq_u8 (uint8x16_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+vmvnq_u16 (uint16x8_t __a)
+{
+  return ~__a;
+}
+
+__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+vmvnq_u32 (uint32x4_t __a)
+{
+  return ~__a;
+}
+
 /* vneg  */
 
 __extension__ static __inline float32x2_t __attribute__ ((__always_inline__))