[ARM][GCC][1/8x]: MVE ACLE vidup, vddup, viwdup and vdwdup intrinsics with writeback.

author Srinath Parvathaneni <srinath.parvathaneni@arm.com>

Fri, 20 Mar 2020 11:58:30 +0000 (11:58 +0000)

committer Kyrylo Tkachov <kyrylo.tkachov@arm.com>

Fri, 20 Mar 2020 11:58:30 +0000 (11:58 +0000)
author Srinath Parvathaneni <srinath.parvathaneni@arm.com>
Fri, 20 Mar 2020 11:58:30 +0000 (11:58 +0000)
committer Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Fri, 20 Mar 2020 11:58:30 +0000 (11:58 +0000)
diff --git a/gcc/ChangeLog b/gcc/ChangeLog

index 55b5b08..22c9766 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,4 +1,147 @@
  2020-03-20  Srinath Parvathaneni  <srinath.parvathaneni@arm.com>
+            Andre Vieira  <andre.simoesdiasvieira@arm.com>
+            Mihail Ionescu  <mihail.ionescu@arm.com>
+
+       * config/arm/arm-builtins.c
+       (QUINOP_UNONE_UNONE_UNONE_UNONE_IMM_UNONE_QUALIFIERS): Define quinary
+       builtin qualifier.
+       * config/arm/arm_mve.h (vddupq_m_n_u8): Define macro.
+       (vddupq_m_n_u32): Likewise.
+       (vddupq_m_n_u16): Likewise.
+       (vddupq_m_wb_u8): Likewise.
+       (vddupq_m_wb_u16): Likewise.
+       (vddupq_m_wb_u32): Likewise.
+       (vddupq_n_u8): Likewise.
+       (vddupq_n_u32): Likewise.
+       (vddupq_n_u16): Likewise.
+       (vddupq_wb_u8): Likewise.
+       (vddupq_wb_u16): Likewise.
+       (vddupq_wb_u32): Likewise.
+       (vdwdupq_m_n_u8): Likewise.
+       (vdwdupq_m_n_u32): Likewise.
+       (vdwdupq_m_n_u16): Likewise.
+       (vdwdupq_m_wb_u8): Likewise.
+       (vdwdupq_m_wb_u32): Likewise.
+       (vdwdupq_m_wb_u16): Likewise.
+       (vdwdupq_n_u8): Likewise.
+       (vdwdupq_n_u32): Likewise.
+       (vdwdupq_n_u16): Likewise.
+       (vdwdupq_wb_u8): Likewise.
+       (vdwdupq_wb_u32): Likewise.
+       (vdwdupq_wb_u16): Likewise.
+       (vidupq_m_n_u8): Likewise.
+       (vidupq_m_n_u32): Likewise.
+       (vidupq_m_n_u16): Likewise.
+       (vidupq_m_wb_u8): Likewise.
+       (vidupq_m_wb_u16): Likewise.
+       (vidupq_m_wb_u32): Likewise.
+       (vidupq_n_u8): Likewise.
+       (vidupq_n_u32): Likewise.
+       (vidupq_n_u16): Likewise.
+       (vidupq_wb_u8): Likewise.
+       (vidupq_wb_u16): Likewise.
+       (vidupq_wb_u32): Likewise.
+       (viwdupq_m_n_u8): Likewise.
+       (viwdupq_m_n_u32): Likewise.
+       (viwdupq_m_n_u16): Likewise.
+       (viwdupq_m_wb_u8): Likewise.
+       (viwdupq_m_wb_u32): Likewise.
+       (viwdupq_m_wb_u16): Likewise.
+       (viwdupq_n_u8): Likewise.
+       (viwdupq_n_u32): Likewise.
+       (viwdupq_n_u16): Likewise.
+       (viwdupq_wb_u8): Likewise.
+       (viwdupq_wb_u32): Likewise.
+       (viwdupq_wb_u16): Likewise.
+       (__arm_vddupq_m_n_u8): Define intrinsic.
+       (__arm_vddupq_m_n_u32): Likewise.
+       (__arm_vddupq_m_n_u16): Likewise.
+       (__arm_vddupq_m_wb_u8): Likewise.
+       (__arm_vddupq_m_wb_u16): Likewise.
+       (__arm_vddupq_m_wb_u32): Likewise.
+       (__arm_vddupq_n_u8): Likewise.
+       (__arm_vddupq_n_u32): Likewise.
+       (__arm_vddupq_n_u16): Likewise.
+       (__arm_vdwdupq_m_n_u8): Likewise.
+       (__arm_vdwdupq_m_n_u32): Likewise.
+       (__arm_vdwdupq_m_n_u16): Likewise.
+       (__arm_vdwdupq_m_wb_u8): Likewise.
+       (__arm_vdwdupq_m_wb_u32): Likewise.
+       (__arm_vdwdupq_m_wb_u16): Likewise.
+       (__arm_vdwdupq_n_u8): Likewise.
+       (__arm_vdwdupq_n_u32): Likewise.
+       (__arm_vdwdupq_n_u16): Likewise.
+       (__arm_vdwdupq_wb_u8): Likewise.
+       (__arm_vdwdupq_wb_u32): Likewise.
+       (__arm_vdwdupq_wb_u16): Likewise.
+       (__arm_vidupq_m_n_u8): Likewise.
+       (__arm_vidupq_m_n_u32): Likewise.
+       (__arm_vidupq_m_n_u16): Likewise.
+       (__arm_vidupq_n_u8): Likewise.
+       (__arm_vidupq_m_wb_u8): Likewise.
+       (__arm_vidupq_m_wb_u16): Likewise.
+       (__arm_vidupq_m_wb_u32): Likewise.
+       (__arm_vidupq_n_u32): Likewise.
+       (__arm_vidupq_n_u16): Likewise.
+       (__arm_vidupq_wb_u8): Likewise.
+       (__arm_vidupq_wb_u16): Likewise.
+       (__arm_vidupq_wb_u32): Likewise.
+       (__arm_vddupq_wb_u8): Likewise.
+       (__arm_vddupq_wb_u16): Likewise.
+       (__arm_vddupq_wb_u32): Likewise.
+       (__arm_viwdupq_m_n_u8): Likewise.
+       (__arm_viwdupq_m_n_u32): Likewise.
+       (__arm_viwdupq_m_n_u16): Likewise.
+       (__arm_viwdupq_m_wb_u8): Likewise.
+       (__arm_viwdupq_m_wb_u32): Likewise.
+       (__arm_viwdupq_m_wb_u16): Likewise.
+       (__arm_viwdupq_n_u8): Likewise.
+       (__arm_viwdupq_n_u32): Likewise.
+       (__arm_viwdupq_n_u16): Likewise.
+       (__arm_viwdupq_wb_u8): Likewise.
+       (__arm_viwdupq_wb_u32): Likewise.
+       (__arm_viwdupq_wb_u16): Likewise.
+       (vidupq_m): Define polymorphic variant.
+       (vddupq_m): Likewise.
+       (vidupq_u16): Likewise.
+       (vidupq_u32): Likewise.
+       (vidupq_u8): Likewise.
+       (vddupq_u16): Likewise.
+       (vddupq_u32): Likewise.
+       (vddupq_u8): Likewise.
+       (viwdupq_m): Likewise.
+       (viwdupq_u16): Likewise.
+       (viwdupq_u32): Likewise.
+       (viwdupq_u8): Likewise.
+       (vdwdupq_m): Likewise.
+       (vdwdupq_u16): Likewise.
+       (vdwdupq_u32): Likewise.
+       (vdwdupq_u8): Likewise.
+       * config/arm/arm_mve_builtins.def
+       (QUINOP_UNONE_UNONE_UNONE_UNONE_IMM_UNONE_QUALIFIERS): Use builtin
+       qualifier.
+       * config/arm/mve.md (mve_vidupq_n_u<mode>): Define RTL pattern.
+       (mve_vidupq_u<mode>_insn): Likewise.
+       (mve_vidupq_m_n_u<mode>): Likewise.
+       (mve_vidupq_m_wb_u<mode>_insn): Likewise.
+       (mve_vddupq_n_u<mode>): Likewise.
+       (mve_vddupq_u<mode>_insn): Likewise.
+       (mve_vddupq_m_n_u<mode>): Likewise.
+       (mve_vddupq_m_wb_u<mode>_insn): Likewise.
+       (mve_vdwdupq_n_u<mode>): Likewise.
+       (mve_vdwdupq_wb_u<mode>): Likewise.
+       (mve_vdwdupq_wb_u<mode>_insn): Likewise.
+       (mve_vdwdupq_m_n_u<mode>): Likewise.
+       (mve_vdwdupq_m_wb_u<mode>): Likewise.
+       (mve_vdwdupq_m_wb_u<mode>_insn): Likewise.
+       (mve_viwdupq_n_u<mode>): Likewise.
+       (mve_viwdupq_wb_u<mode>): Likewise.
+       (mve_viwdupq_wb_u<mode>_insn): Likewise.
+       (mve_viwdupq_m_n_u<mode>): Likewise.
+       (mve_viwdupq_m_wb_u<mode>): Likewise.
+       (mve_viwdupq_m_wb_u<mode>_insn): Likewise.
+
+2020-03-20  Srinath Parvathaneni  <srinath.parvathaneni@arm.com>
  
         * config/arm/arm_mve.h (vreinterpretq_s16_s32): Define macro.
         (vreinterpretq_s16_s64): Likewise.
diff --git a/gcc/config/arm/arm-builtins.c b/gcc/config/arm/arm-builtins.c

index c3deb9e..cefc144 100644 (file)
--- a/gcc/config/arm/arm-builtins.c
+++ b/gcc/config/arm/arm-builtins.c
@@ -711,6 +711,13 @@ arm_ldru_z_qualifiers[SIMD_MAX_BUILTIN_ARGS]
    = { qualifier_unsigned, qualifier_pointer, qualifier_unsigned};
  #define LDRU_Z_QUALIFIERS (arm_ldru_z_qualifiers)
  
+static enum arm_type_qualifiers
+arm_quinop_unone_unone_unone_unone_imm_unone_qualifiers[SIMD_MAX_BUILTIN_ARGS]
+  = { qualifier_unsigned, qualifier_unsigned, qualifier_unsigned,
+      qualifier_unsigned, qualifier_immediate, qualifier_unsigned };
+#define QUINOP_UNONE_UNONE_UNONE_UNONE_IMM_UNONE_QUALIFIERS \
+  (arm_quinop_unone_unone_unone_unone_imm_unone_qualifiers)
+
  /* End of Qualifier for MVE builtins.  */
  
     /* void ([T element type] *, T, immediate).  */
diff --git a/gcc/config/arm/arm_mve.h b/gcc/config/arm/arm_mve.h

index 916565c..00f2242 100644 (file)
--- a/gcc/config/arm/arm_mve.h
+++ b/gcc/config/arm/arm_mve.h
@@ -2006,6 +2006,54 @@ typedef struct { uint8x16_t val[4]; } uint8x16x4_t;
  #define vuninitializedq_s64(void) __arm_vuninitializedq_s64(void)
  #define vuninitializedq_f16(void) __arm_vuninitializedq_f16(void)
  #define vuninitializedq_f32(void) __arm_vuninitializedq_f32(void)
+#define vddupq_m_n_u8(__inactive, __a,  __imm, __p) __arm_vddupq_m_n_u8(__inactive, __a,  __imm, __p)
+#define vddupq_m_n_u32(__inactive, __a,  __imm, __p) __arm_vddupq_m_n_u32(__inactive, __a,  __imm, __p)
+#define vddupq_m_n_u16(__inactive, __a,  __imm, __p) __arm_vddupq_m_n_u16(__inactive, __a,  __imm, __p)
+#define vddupq_m_wb_u8(__inactive,  __a,  __imm, __p) __arm_vddupq_m_wb_u8(__inactive,  __a,  __imm, __p)
+#define vddupq_m_wb_u16(__inactive,  __a,  __imm, __p) __arm_vddupq_m_wb_u16(__inactive,  __a,  __imm, __p)
+#define vddupq_m_wb_u32(__inactive,  __a,  __imm, __p) __arm_vddupq_m_wb_u32(__inactive,  __a,  __imm, __p)
+#define vddupq_n_u8(__a,  __imm) __arm_vddupq_n_u8(__a,  __imm)
+#define vddupq_n_u32(__a,  __imm) __arm_vddupq_n_u32(__a,  __imm)
+#define vddupq_n_u16(__a,  __imm) __arm_vddupq_n_u16(__a,  __imm)
+#define vddupq_wb_u8( __a,  __imm) __arm_vddupq_wb_u8( __a,  __imm)
+#define vddupq_wb_u16( __a,  __imm) __arm_vddupq_wb_u16( __a,  __imm)
+#define vddupq_wb_u32( __a,  __imm) __arm_vddupq_wb_u32( __a,  __imm)
+#define vdwdupq_m_n_u8(__inactive, __a, __b,  __imm, __p) __arm_vdwdupq_m_n_u8(__inactive, __a, __b,  __imm, __p)
+#define vdwdupq_m_n_u32(__inactive, __a, __b,  __imm, __p) __arm_vdwdupq_m_n_u32(__inactive, __a, __b,  __imm, __p)
+#define vdwdupq_m_n_u16(__inactive, __a, __b,  __imm, __p) __arm_vdwdupq_m_n_u16(__inactive, __a, __b,  __imm, __p)
+#define vdwdupq_m_wb_u8(__inactive,  __a, __b,  __imm, __p) __arm_vdwdupq_m_wb_u8(__inactive,  __a, __b,  __imm, __p)
+#define vdwdupq_m_wb_u32(__inactive,  __a, __b,  __imm, __p) __arm_vdwdupq_m_wb_u32(__inactive,  __a, __b,  __imm, __p)
+#define vdwdupq_m_wb_u16(__inactive,  __a, __b,  __imm, __p) __arm_vdwdupq_m_wb_u16(__inactive,  __a, __b,  __imm, __p)
+#define vdwdupq_n_u8(__a, __b,  __imm) __arm_vdwdupq_n_u8(__a, __b,  __imm)
+#define vdwdupq_n_u32(__a, __b,  __imm) __arm_vdwdupq_n_u32(__a, __b,  __imm)
+#define vdwdupq_n_u16(__a, __b,  __imm) __arm_vdwdupq_n_u16(__a, __b,  __imm)
+#define vdwdupq_wb_u8( __a, __b,  __imm) __arm_vdwdupq_wb_u8( __a, __b,  __imm)
+#define vdwdupq_wb_u32( __a, __b,  __imm) __arm_vdwdupq_wb_u32( __a, __b,  __imm)
+#define vdwdupq_wb_u16( __a, __b,  __imm) __arm_vdwdupq_wb_u16( __a, __b,  __imm)
+#define vidupq_m_n_u8(__inactive, __a,  __imm, __p) __arm_vidupq_m_n_u8(__inactive, __a,  __imm, __p)
+#define vidupq_m_n_u32(__inactive, __a,  __imm, __p) __arm_vidupq_m_n_u32(__inactive, __a,  __imm, __p)
+#define vidupq_m_n_u16(__inactive, __a,  __imm, __p) __arm_vidupq_m_n_u16(__inactive, __a,  __imm, __p)
+#define vidupq_m_wb_u8(__inactive,  __a,  __imm, __p) __arm_vidupq_m_wb_u8(__inactive,  __a,  __imm, __p)
+#define vidupq_m_wb_u16(__inactive,  __a,  __imm, __p) __arm_vidupq_m_wb_u16(__inactive,  __a,  __imm, __p)
+#define vidupq_m_wb_u32(__inactive,  __a,  __imm, __p) __arm_vidupq_m_wb_u32(__inactive,  __a,  __imm, __p)
+#define vidupq_n_u8(__a,  __imm) __arm_vidupq_n_u8(__a,  __imm)
+#define vidupq_n_u32(__a,  __imm) __arm_vidupq_n_u32(__a,  __imm)
+#define vidupq_n_u16(__a,  __imm) __arm_vidupq_n_u16(__a,  __imm)
+#define vidupq_wb_u8( __a,  __imm) __arm_vidupq_wb_u8( __a,  __imm)
+#define vidupq_wb_u16( __a,  __imm) __arm_vidupq_wb_u16( __a,  __imm)
+#define vidupq_wb_u32( __a,  __imm) __arm_vidupq_wb_u32( __a,  __imm)
+#define viwdupq_m_n_u8(__inactive, __a, __b,  __imm, __p) __arm_viwdupq_m_n_u8(__inactive, __a, __b,  __imm, __p)
+#define viwdupq_m_n_u32(__inactive, __a, __b,  __imm, __p) __arm_viwdupq_m_n_u32(__inactive, __a, __b,  __imm, __p)
+#define viwdupq_m_n_u16(__inactive, __a, __b,  __imm, __p) __arm_viwdupq_m_n_u16(__inactive, __a, __b,  __imm, __p)
+#define viwdupq_m_wb_u8(__inactive,  __a, __b,  __imm, __p) __arm_viwdupq_m_wb_u8(__inactive,  __a, __b,  __imm, __p)
+#define viwdupq_m_wb_u32(__inactive,  __a, __b,  __imm, __p) __arm_viwdupq_m_wb_u32(__inactive,  __a, __b,  __imm, __p)
+#define viwdupq_m_wb_u16(__inactive,  __a, __b,  __imm, __p) __arm_viwdupq_m_wb_u16(__inactive,  __a, __b,  __imm, __p)
+#define viwdupq_n_u8(__a, __b,  __imm) __arm_viwdupq_n_u8(__a, __b,  __imm)
+#define viwdupq_n_u32(__a, __b,  __imm) __arm_viwdupq_n_u32(__a, __b,  __imm)
+#define viwdupq_n_u16(__a, __b,  __imm) __arm_viwdupq_n_u16(__a, __b,  __imm)
+#define viwdupq_wb_u8( __a, __b,  __imm) __arm_viwdupq_wb_u8( __a, __b,  __imm)
+#define viwdupq_wb_u32( __a, __b,  __imm) __arm_viwdupq_wb_u32( __a, __b,  __imm)
+#define viwdupq_wb_u16( __a, __b,  __imm) __arm_viwdupq_wb_u16( __a, __b,  __imm)
  #endif
  
  __extension__ extern __inline void
@@ -12956,6 +13004,390 @@ __arm_vreinterpretq_u8_u64 (uint64x2_t __a)
    return (uint8x16_t)  __a;
  }
  
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_m_n_u8 (uint8x16_t __inactive, uint32_t __a, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vddupq_m_n_uv16qi (__inactive, __a, __imm, __p);
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_m_n_u32 (uint32x4_t __inactive, uint32_t __a, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vddupq_m_n_uv4si (__inactive, __a, __imm, __p);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_m_n_u16 (uint16x8_t __inactive, uint32_t __a, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vddupq_m_n_uv8hi (__inactive, __a, __imm, __p);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_m_wb_u8 (uint8x16_t __inactive, uint32_t * __a, const int __imm, mve_pred16_t __p)
+{
+  uint8x16_t __res = __builtin_mve_vddupq_m_n_uv16qi (__inactive, * __a, __imm, __p);
+  *__a -= __imm * 16u;
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_m_wb_u16 (uint16x8_t __inactive, uint32_t * __a, const int __imm, mve_pred16_t __p)
+{
+  uint16x8_t __res = __builtin_mve_vddupq_m_n_uv8hi (__inactive, *__a, __imm, __p);
+  *__a -= __imm * 8u;
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_m_wb_u32 (uint32x4_t __inactive, uint32_t * __a, const int __imm, mve_pred16_t __p)
+{
+  uint32x4_t __res = __builtin_mve_vddupq_m_n_uv4si (__inactive, *__a, __imm, __p);
+  *__a -= __imm * 4u;
+  return __res;
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_n_u8 (uint32_t __a, const int __imm)
+{
+  return __builtin_mve_vddupq_n_uv16qi (__a, __imm);
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_n_u32 (uint32_t __a, const int __imm)
+{
+  return __builtin_mve_vddupq_n_uv4si (__a, __imm);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_n_u16 (uint32_t __a, const int __imm)
+{
+  return __builtin_mve_vddupq_n_uv8hi (__a, __imm);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_m_n_u8 (uint8x16_t __inactive, uint32_t __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vdwdupq_m_n_uv16qi (__inactive, __a, __b, __imm, __p);
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_m_n_u32 (uint32x4_t __inactive, uint32_t __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vdwdupq_m_n_uv4si (__inactive, __a, __b, __imm, __p);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_m_n_u16 (uint16x8_t __inactive, uint32_t __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vdwdupq_m_n_uv8hi (__inactive, __a, __b, __imm, __p);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_m_wb_u8 (uint8x16_t __inactive, uint32_t * __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  uint8x16_t __res =  __builtin_mve_vdwdupq_m_n_uv16qi (__inactive, *__a, __b, __imm, __p);
+  *__a = __builtin_mve_vdwdupq_m_wb_uv16qi (__inactive, *__a, __b, __imm, __p);
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_m_wb_u32 (uint32x4_t __inactive, uint32_t * __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  uint32x4_t __res =  __builtin_mve_vdwdupq_m_n_uv4si (__inactive, *__a, __b, __imm, __p);
+  *__a = __builtin_mve_vdwdupq_m_wb_uv4si (__inactive, *__a, __b, __imm, __p);
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_m_wb_u16 (uint16x8_t __inactive, uint32_t * __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  uint16x8_t __res =  __builtin_mve_vdwdupq_m_n_uv8hi (__inactive, *__a, __b, __imm, __p);
+  *__a = __builtin_mve_vdwdupq_m_wb_uv8hi (__inactive, *__a, __b, __imm, __p);
+  return __res;
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_n_u8 (uint32_t __a, uint32_t __b, const int __imm)
+{
+  return __builtin_mve_vdwdupq_n_uv16qi (__a, __b, __imm);
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_n_u32 (uint32_t __a, uint32_t __b, const int __imm)
+{
+  return __builtin_mve_vdwdupq_n_uv4si (__a, __b, __imm);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_n_u16 (uint32_t __a, uint32_t __b, const int __imm)
+{
+  return __builtin_mve_vdwdupq_n_uv8hi (__a, __b, __imm);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_wb_u8 (uint32_t * __a, uint32_t __b, const int __imm)
+{
+  uint8x16_t __res = __builtin_mve_vdwdupq_n_uv16qi (*__a, __b, __imm);
+  *__a = __builtin_mve_vdwdupq_wb_uv16qi (*__a, __b, __imm);
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_wb_u32 (uint32_t * __a, uint32_t __b, const int __imm)
+{
+  uint32x4_t __res = __builtin_mve_vdwdupq_n_uv4si (*__a, __b, __imm);
+  *__a = __builtin_mve_vdwdupq_wb_uv4si (*__a, __b, __imm);
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vdwdupq_wb_u16 (uint32_t * __a, uint32_t __b, const int __imm)
+{
+  uint16x8_t __res = __builtin_mve_vdwdupq_n_uv8hi (*__a, __b, __imm);
+  *__a = __builtin_mve_vdwdupq_wb_uv8hi (*__a, __b, __imm);
+  return __res;
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_m_n_u8 (uint8x16_t __inactive, uint32_t __a, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vidupq_m_n_uv16qi (__inactive, __a, __imm, __p);
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_m_n_u32 (uint32x4_t __inactive, uint32_t __a, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vidupq_m_n_uv4si (__inactive, __a, __imm, __p);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_m_n_u16 (uint16x8_t __inactive, uint32_t __a, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_vidupq_m_n_uv8hi (__inactive, __a, __imm, __p);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_n_u8 (uint32_t __a, const int __imm)
+{
+  return __builtin_mve_vidupq_n_uv16qi (__a, __imm);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_m_wb_u8 (uint8x16_t __inactive, uint32_t * __a, const int __imm, mve_pred16_t __p)
+{
+  uint8x16_t __res = __builtin_mve_vidupq_m_n_uv16qi (__inactive, *__a, __imm, __p);
+  *__a += __imm * 16u;
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_m_wb_u16 (uint16x8_t __inactive, uint32_t * __a, const int __imm, mve_pred16_t __p)
+{
+  uint16x8_t __res = __builtin_mve_vidupq_m_n_uv8hi (__inactive, *__a, __imm, __p);
+  *__a += __imm * 8u;
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_m_wb_u32 (uint32x4_t __inactive, uint32_t * __a, const int __imm, mve_pred16_t __p)
+{
+  uint32x4_t __res = __builtin_mve_vidupq_m_n_uv4si (__inactive, *__a, __imm, __p);
+  *__a += __imm * 4u;
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_n_u32 (uint32_t __a, const int __imm)
+{
+  return __builtin_mve_vidupq_n_uv4si (__a, __imm);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_n_u16 (uint32_t __a, const int __imm)
+{
+  return __builtin_mve_vidupq_n_uv8hi (__a, __imm);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_wb_u8 (uint32_t * __a, const int __imm)
+{
+  uint8x16_t __res = __builtin_mve_vidupq_n_uv16qi (*__a, __imm);
+  *__a += __imm * 16u;
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_wb_u16 (uint32_t * __a, const int __imm)
+{
+  uint16x8_t __res = __builtin_mve_vidupq_n_uv8hi (*__a, __imm);
+  *__a += __imm * 8u;
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vidupq_wb_u32 (uint32_t * __a, const int __imm)
+{
+  uint32x4_t __res = __builtin_mve_vidupq_n_uv4si (*__a, __imm);
+  *__a += __imm * 4u;
+  return __res;
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_wb_u8 (uint32_t * __a, const int __imm)
+{
+  uint8x16_t __res = __builtin_mve_vddupq_n_uv16qi (*__a, __imm);
+  *__a -= __imm * 16u;
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_wb_u16 (uint32_t * __a, const int __imm)
+{
+  uint16x8_t __res = __builtin_mve_vddupq_n_uv8hi (*__a, __imm);
+  *__a -= __imm * 8u;
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vddupq_wb_u32 (uint32_t * __a, const int __imm)
+{
+  uint32x4_t __res = __builtin_mve_vddupq_n_uv4si (*__a, __imm);
+  *__a -= __imm * 4u;
+  return __res;
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_m_n_u8 (uint8x16_t __inactive, uint32_t __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_viwdupq_m_n_uv16qi (__inactive, __a, __b, __imm, __p);
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_m_n_u32 (uint32x4_t __inactive, uint32_t __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_viwdupq_m_n_uv4si (__inactive, __a, __b, __imm, __p);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_m_n_u16 (uint16x8_t __inactive, uint32_t __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  return __builtin_mve_viwdupq_m_n_uv8hi (__inactive, __a, __b, __imm, __p);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_m_wb_u8 (uint8x16_t __inactive, uint32_t * __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  uint8x16_t __res = __builtin_mve_viwdupq_m_n_uv16qi (__inactive, *__a, __b, __imm, __p);
+  *__a =  __builtin_mve_viwdupq_m_wb_uv16qi (__inactive, *__a, __b, __imm, __p);
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_m_wb_u32 (uint32x4_t __inactive, uint32_t * __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  uint32x4_t __res = __builtin_mve_viwdupq_m_n_uv4si (__inactive, *__a, __b, __imm, __p);
+  *__a =  __builtin_mve_viwdupq_m_wb_uv4si (__inactive, *__a, __b, __imm, __p);
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_m_wb_u16 (uint16x8_t __inactive, uint32_t * __a, uint32_t __b, const int __imm, mve_pred16_t __p)
+{
+  uint16x8_t __res = __builtin_mve_viwdupq_m_n_uv8hi (__inactive, *__a, __b, __imm, __p);
+  *__a =  __builtin_mve_viwdupq_m_wb_uv8hi (__inactive, *__a, __b, __imm, __p);
+  return __res;
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_n_u8 (uint32_t __a, uint32_t __b, const int __imm)
+{
+  return __builtin_mve_viwdupq_n_uv16qi (__a, __b, __imm);
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_n_u32 (uint32_t __a, uint32_t __b, const int __imm)
+{
+  return __builtin_mve_viwdupq_n_uv4si (__a, __b, __imm);
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_n_u16 (uint32_t __a, uint32_t __b, const int __imm)
+{
+  return __builtin_mve_viwdupq_n_uv8hi (__a, __b, __imm);
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_wb_u8 (uint32_t * __a, uint32_t __b, const int __imm)
+{
+  uint8x16_t __res = __builtin_mve_viwdupq_n_uv16qi (*__a, __b, __imm);
+  *__a = __builtin_mve_viwdupq_wb_uv16qi (*__a, __b, __imm);
+  return __res;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_wb_u32 (uint32_t * __a, uint32_t __b, const int __imm)
+{
+  uint32x4_t __res = __builtin_mve_viwdupq_n_uv4si (*__a, __b, __imm);
+  *__a = __builtin_mve_viwdupq_wb_uv4si (*__a, __b, __imm);
+  return __res;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_viwdupq_wb_u16 (uint32_t * __a, uint32_t __b, const int __imm)
+{
+  uint16x8_t __res = __builtin_mve_viwdupq_n_uv8hi (*__a, __b, __imm);
+  *__a = __builtin_mve_viwdupq_wb_uv8hi (*__a, __b, __imm);
+  return __res;
+}
+
  #if (__ARM_FEATURE_MVE & 2) /* MVE Floating point.  */
  
  __extension__ extern __inline void
@@ -21764,6 +22196,122 @@ extern void *__ARM_undef;
    int (*)[__ARM_mve_type_uint8_t_const_ptr][__ARM_mve_type_uint16x8_t]: __arm_vldrbq_gather_offset_u16 (__ARM_mve_coerce(__p0, uint8_t const *), __ARM_mve_coerce(__p1, uint16x8_t)), \
    int (*)[__ARM_mve_type_uint8_t_const_ptr][__ARM_mve_type_uint32x4_t]: __arm_vldrbq_gather_offset_u32 (__ARM_mve_coerce(__p0, uint8_t const *), __ARM_mve_coerce(__p1, uint32x4_t)));})
  
+#define vidupq_m(p0,p1,p2,p3) __arm_vidupq_m(p0,p1,p2,p3)
+#define __arm_vidupq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
+ __typeof(p1) __p1 = (p1); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
+ int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t]: __arm_vidupq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3), \
+ int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t]: __arm_vidupq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3), \
+ int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t]: __arm_vidupq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3), \
+ int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t_ptr]: __arm_vidupq_m_wb_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3), \
+ int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t_ptr]: __arm_vidupq_m_wb_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3), \
+ int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t_ptr]: __arm_vidupq_m_wb_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3));})
+
+#define vddupq_m(p0,p1,p2,p3) __arm_vddupq_m(p0,p1,p2,p3)
+#define __arm_vddupq_m(p0,p1,p2,p3) ({ __typeof(p0) __p0 = (p0); \
+ __typeof(p1) __p1 = (p1); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
+ int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t]: __arm_vddupq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3), \
+ int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t]: __arm_vddupq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3), \
+ int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t]: __arm_vddupq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3), \
+ int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t_ptr]: __arm_vddupq_m_wb_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3), \
+ int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t_ptr]: __arm_vddupq_m_wb_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3), \
+ int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t_ptr]: __arm_vddupq_m_wb_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3));})
+
+#define vidupq_u16(p0,p1) __arm_vidupq_u16(p0,p1)
+#define __arm_vidupq_u16(p0,p1) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vidupq_n_u16 (__ARM_mve_coerce(__p0, uint32_t), p1), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vidupq_wb_u16 (__ARM_mve_coerce(__p0, uint32_t *), p1));})
+
+#define vidupq_u32(p0,p1) __arm_vidupq_u32(p0,p1)
+#define __arm_vidupq_u32(p0,p1) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vidupq_n_u32 (__ARM_mve_coerce(__p0, uint32_t), p1), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vidupq_wb_u32 (__ARM_mve_coerce(__p0, uint32_t *), p1));})
+
+#define vidupq_u8(p0,p1) __arm_vidupq_u8(p0,p1)
+#define __arm_vidupq_u8(p0,p1) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vidupq_n_u8 (__ARM_mve_coerce(__p0, uint32_t), p1), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vidupq_wb_u8 (__ARM_mve_coerce(__p0, uint32_t *), p1));})
+
+#define vddupq_u16(p0,p1) __arm_vddupq_u16(p0,p1)
+#define __arm_vddupq_u16(p0,p1) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vddupq_n_u16 (__ARM_mve_coerce(__p0, uint32_t), p1), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vddupq_wb_u16 (__ARM_mve_coerce(__p0, uint32_t *), p1));})
+
+#define vddupq_u32(p0,p1) __arm_vddupq_u32(p0,p1)
+#define __arm_vddupq_u32(p0,p1) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vddupq_n_u32 (__ARM_mve_coerce(__p0, uint32_t), p1), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vddupq_wb_u32 (__ARM_mve_coerce(__p0, uint32_t *), p1));})
+
+#define vddupq_u8(p0,p1) __arm_vddupq_u8(p0,p1)
+#define __arm_vddupq_u8(p0,p1) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vddupq_n_u8 (__ARM_mve_coerce(__p0, uint32_t), p1), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vddupq_wb_u8 (__ARM_mve_coerce(__p0, uint32_t *), p1));})
+
+#define viwdupq_m(p0,p1,p2,p3,p4) __arm_viwdupq_m(p0,p1,p2,p3,p4)
+#define __arm_viwdupq_m(p0,p1,p2,p3,p4) ({ __typeof(p0) __p0 = (p0); \
+  __typeof(p1) __p1 = (p1); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
+  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t]: __arm_viwdupq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t]: __arm_viwdupq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t]: __arm_viwdupq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t_ptr]: __arm_viwdupq_m_wb_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t_ptr]: __arm_viwdupq_m_wb_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t_ptr]: __arm_viwdupq_m_wb_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3, p4));})
+
+#define viwdupq_u16(p0,p1,p2) __arm_viwdupq_u16(p0,p1,p2)
+#define __arm_viwdupq_u16(p0,p1,p2) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_viwdupq_n_u16 (__ARM_mve_coerce(__p0, uint32_t), p1, p2), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_viwdupq_wb_u16 (__ARM_mve_coerce(__p0, uint32_t *), p1, p2));})
+
+#define viwdupq_u32(p0,p1,p2) __arm_viwdupq_u32(p0,p1,p2)
+#define __arm_viwdupq_u32(p0,p1,p2) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_viwdupq_n_u32 (__ARM_mve_coerce(__p0, uint32_t), p1, p2), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_viwdupq_wb_u32 (__ARM_mve_coerce(__p0, uint32_t *), p1, p2));})
+
+#define viwdupq_u8(p0,p1,p2) __arm_viwdupq_u8(p0,p1,p2)
+#define __arm_viwdupq_u8(p0,p1,p2) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_viwdupq_n_u8 (__ARM_mve_coerce(__p0, uint32_t), p1, p2), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_viwdupq_wb_u8 (__ARM_mve_coerce(__p0, uint32_t *), p1, p2));})
+
+#define vdwdupq_m(p0,p1,p2,p3,p4) __arm_vdwdupq_m(p0,p1,p2,p3,p4)
+#define __arm_vdwdupq_m(p0,p1,p2,p3,p4) ({ __typeof(p0) __p0 = (p0); \
+  __typeof(p1) __p1 = (p1); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
+  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t]: __arm_vdwdupq_m_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t]: __arm_vdwdupq_m_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t]: __arm_vdwdupq_m_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint32_t_ptr]: __arm_vdwdupq_m_wb_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint32_t_ptr]: __arm_vdwdupq_m_wb_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3, p4), \
+  int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t_ptr]: __arm_vdwdupq_m_wb_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t *), p2, p3, p4));})
+
+#define vdwdupq_u16(p0,p1,p2) __arm_vdwdupq_u16(p0,p1,p2)
+#define __arm_vdwdupq_u16(p0,p1,p2) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vdwdupq_n_u16 (__ARM_mve_coerce(__p0, uint32_t), p1, p2), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vdwdupq_wb_u16 (__ARM_mve_coerce(__p0, uint32_t *), p1, p2));})
+
+#define vdwdupq_u32(p0,p1,p2) __arm_vdwdupq_u32(p0,p1,p2)
+#define __arm_vdwdupq_u32(p0,p1,p2) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vdwdupq_n_u32 (__ARM_mve_coerce(__p0, uint32_t), p1, p2), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vdwdupq_wb_u32 (__ARM_mve_coerce(__p0, uint32_t *), p1, p2));})
+
+#define vdwdupq_u8(p0,p1,p2) __arm_vdwdupq_u8(p0,p1,p2)
+#define __arm_vdwdupq_u8(p0,p1,p2) ({ __typeof(p0) __p0 = (p0); \
+  _Generic( (int (*)[__ARM_mve_typeid(__p0)])0, \
+  int (*)[__ARM_mve_type_uint32_t]: __arm_vdwdupq_n_u8 (__ARM_mve_coerce(__p0, uint32_t), p1, p2), \
+  int (*)[__ARM_mve_type_uint32_t_ptr]: __arm_vdwdupq_wb_u8 (__ARM_mve_coerce(__p0, uint32_t *), p1, p2));})
+
  #ifdef __cplusplus
  }
  #endif
diff --git a/gcc/config/arm/arm_mve_builtins.def b/gcc/config/arm/arm_mve_builtins.def

index 144547f..2ed7886 100644 (file)
--- a/gcc/config/arm/arm_mve_builtins.def
+++ b/gcc/config/arm/arm_mve_builtins.def
@@ -815,3 +815,15 @@ VAR1 (STRSU_P, vstrdq_scatter_offset_p_u, v2di)
  VAR1 (STRSU_P, vstrdq_scatter_shifted_offset_p_u, v2di)
  VAR1 (STRSU_P, vstrwq_scatter_offset_p_u, v4si)
  VAR1 (STRSU_P, vstrwq_scatter_shifted_offset_p_u, v4si)
+VAR3 (TERNOP_UNONE_UNONE_UNONE_IMM, viwdupq_wb_u, v16qi, v4si, v8hi)
+VAR3 (TERNOP_UNONE_UNONE_UNONE_IMM, vdwdupq_wb_u, v16qi, v4si, v8hi)
+VAR3 (QUINOP_UNONE_UNONE_UNONE_UNONE_IMM_UNONE, viwdupq_m_wb_u, v16qi, v8hi, v4si)
+VAR3 (QUINOP_UNONE_UNONE_UNONE_UNONE_IMM_UNONE, vdwdupq_m_wb_u, v16qi, v8hi, v4si)
+VAR3 (QUINOP_UNONE_UNONE_UNONE_UNONE_IMM_UNONE, viwdupq_m_n_u, v16qi, v8hi, v4si)
+VAR3 (QUINOP_UNONE_UNONE_UNONE_UNONE_IMM_UNONE, vdwdupq_m_n_u, v16qi, v8hi, v4si)
+VAR3 (BINOP_UNONE_UNONE_IMM, vddupq_n_u, v16qi, v8hi, v4si)
+VAR3 (BINOP_UNONE_UNONE_IMM, vidupq_n_u, v16qi, v8hi, v4si)
+VAR3 (QUADOP_UNONE_UNONE_UNONE_IMM_UNONE, vddupq_m_n_u, v16qi, v8hi, v4si)
+VAR3 (QUADOP_UNONE_UNONE_UNONE_IMM_UNONE, vidupq_m_n_u, v16qi, v8hi, v4si)
+VAR3 (TERNOP_UNONE_UNONE_UNONE_IMM, vdwdupq_n_u, v16qi, v4si, v8hi)
+VAR3 (TERNOP_UNONE_UNONE_UNONE_IMM, viwdupq_n_u, v16qi, v4si, v8hi)
diff --git a/gcc/config/arm/mve.md b/gcc/config/arm/mve.md

index 77b36a7..b2702f5 100644 (file)
--- a/gcc/config/arm/mve.md
+++ b/gcc/config/arm/mve.md
@@ -207,7 +207,8 @@
                          VSTRDQSB_U VSTRDQSO_S VSTRDQSO_U VSTRDQSSO_S
                          VSTRDQSSO_U VSTRWQSO_S VSTRWQSO_U VSTRWQSSO_S
                          VSTRWQSSO_U VSTRHQSO_F VSTRHQSSO_F VSTRWQSB_F
-                        VSTRWQSO_F VSTRWQSSO_F])
+                        VSTRWQSO_F VSTRWQSSO_F VDDUPQ VDDUPQ_M VDWDUPQ
+                        VDWDUPQ_M VIDUPQ VIDUPQ_M VIWDUPQ VIWDUPQ_M])
  
  (define_mode_attr MVE_CNVT [(V8HI "V8HF") (V4SI "V4SF") (V8HF "V8HI")
                             (V4SF "V4SI")])
@@ -9671,3 +9672,373 @@
    "vadd.f%#<V_sz_elem> %q0, %q1, %q2"
    [(set_attr "type" "mve_move")
  ])
+
+;;
+;; [vidupq_n_u])
+;;
+(define_expand "mve_vidupq_n_u<mode>"
+ [(match_operand:MVE_2 0 "s_register_operand")
+  (match_operand:SI 1 "s_register_operand")
+  (match_operand:SI 2 "mve_imm_selective_upto_8")]
+ "TARGET_HAVE_MVE"
+{
+  rtx temp = gen_reg_rtx (SImode);
+  emit_move_insn (temp, operands[1]);
+  rtx inc = gen_int_mode (INTVAL(operands[2]) * <MVE_LANES>, SImode);
+  emit_insn (gen_mve_vidupq_u<mode>_insn (operands[0], temp, operands[1],
+                                         operands[2], inc));
+  DONE;
+})
+
+;;
+;; [vidupq_u_insn])
+;;
+(define_insn "mve_vidupq_u<mode>_insn"
+ [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:SI 2 "s_register_operand" "1")
+                     (match_operand:SI 3 "mve_imm_selective_upto_8" "Rg")]
+        VIDUPQ))
+  (set (match_operand:SI 1 "s_register_operand" "=e")
+       (plus:SI (match_dup 2)
+               (match_operand:SI 4 "immediate_operand" "i")))]
+ "TARGET_HAVE_MVE"
+ "vidup.u%#<V_sz_elem>\t%q0, %1, %3")
+
+;;
+;; [vidupq_m_n_u])
+;;
+(define_expand "mve_vidupq_m_n_u<mode>"
+  [(match_operand:MVE_2 0 "s_register_operand")
+   (match_operand:MVE_2 1 "s_register_operand")
+   (match_operand:SI 2 "s_register_operand")
+   (match_operand:SI 3 "mve_imm_selective_upto_8")
+   (match_operand:HI 4 "vpr_register_operand")]
+  "TARGET_HAVE_MVE"
+{
+  rtx temp = gen_reg_rtx (SImode);
+  emit_move_insn (temp, operands[2]);
+  rtx inc = gen_int_mode (INTVAL(operands[3]) * <MVE_LANES>, SImode);
+  emit_insn (gen_mve_vidupq_m_wb_u<mode>_insn(operands[0], operands[1], temp,
+                                            operands[2], operands[3],
+                                            operands[4], inc));
+  DONE;
+})
+
+;;
+;; [vidupq_m_wb_u_insn])
+;;
+(define_insn "mve_vidupq_m_wb_u<mode>_insn"
+ [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:MVE_2 1 "s_register_operand" "0")
+                     (match_operand:SI 3 "s_register_operand" "2")
+                     (match_operand:SI 4 "mve_imm_selective_upto_8" "Rg")
+                     (match_operand:HI 5 "vpr_register_operand" "Up")]
+       VIDUPQ_M))
+  (set (match_operand:SI 2 "s_register_operand" "=e")
+       (plus:SI (match_dup 3)
+               (match_operand:SI 6 "immediate_operand" "i")))]
+ "TARGET_HAVE_MVE"
+ "vpst\;\tvidupt.u%#<V_sz_elem>\t%q0, %2, %4"
+ [(set_attr "length""8")])
+
+;;
+;; [vddupq_n_u])
+;;
+(define_expand "mve_vddupq_n_u<mode>"
+ [(match_operand:MVE_2 0 "s_register_operand")
+  (match_operand:SI 1 "s_register_operand")
+  (match_operand:SI 2 "mve_imm_selective_upto_8")]
+ "TARGET_HAVE_MVE"
+{
+  rtx temp = gen_reg_rtx (SImode);
+  emit_move_insn (temp, operands[1]);
+  rtx inc = gen_int_mode (INTVAL(operands[2]) * <MVE_LANES>, SImode);
+  emit_insn (gen_mve_vddupq_u<mode>_insn (operands[0], temp, operands[1],
+                                         operands[2], inc));
+  DONE;
+})
+
+;;
+;; [vddupq_u_insn])
+;;
+(define_insn "mve_vddupq_u<mode>_insn"
+ [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:SI 2 "s_register_operand" "1")
+                     (match_operand:SI 3 "immediate_operand" "i")]
+       VDDUPQ))
+  (set (match_operand:SI 1 "s_register_operand" "=e")
+       (minus:SI (match_dup 2)
+                (match_operand:SI 4 "immediate_operand" "i")))]
+ "TARGET_HAVE_MVE"
+ "vddup.u%#<V_sz_elem>  %q0, %1, %3")
+
+;;
+;; [vddupq_m_n_u])
+;;
+(define_expand "mve_vddupq_m_n_u<mode>"
+  [(match_operand:MVE_2 0 "s_register_operand")
+   (match_operand:MVE_2 1 "s_register_operand")
+   (match_operand:SI 2 "s_register_operand")
+   (match_operand:SI 3 "mve_imm_selective_upto_8")
+   (match_operand:HI 4 "vpr_register_operand")]
+  "TARGET_HAVE_MVE"
+{
+  rtx temp = gen_reg_rtx (SImode);
+  emit_move_insn (temp, operands[2]);
+  rtx inc = gen_int_mode (INTVAL(operands[3]) * <MVE_LANES>, SImode);
+  emit_insn (gen_mve_vddupq_m_wb_u<mode>_insn(operands[0], operands[1], temp,
+                                            operands[2], operands[3],
+                                            operands[4], inc));
+  DONE;
+})
+
+;;
+;; [vddupq_m_wb_u_insn])
+;;
+(define_insn "mve_vddupq_m_wb_u<mode>_insn"
+ [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:MVE_2 1 "s_register_operand" "0")
+                     (match_operand:SI 3 "s_register_operand" "2")
+                     (match_operand:SI 4 "mve_imm_selective_upto_8" "Rg")
+                     (match_operand:HI 5 "vpr_register_operand" "Up")]
+       VDDUPQ_M))
+  (set (match_operand:SI 2 "s_register_operand" "=e")
+       (minus:SI (match_dup 3)
+                (match_operand:SI 6 "immediate_operand" "i")))]
+ "TARGET_HAVE_MVE"
+ "vpst\;\tvddupt.u%#<V_sz_elem>\t%q0, %2, %4"
+ [(set_attr "length""8")])
+
+;;
+;; [vdwdupq_n_u])
+;;
+(define_expand "mve_vdwdupq_n_u<mode>"
+ [(match_operand:MVE_2 0 "s_register_operand")
+  (match_operand:SI 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "mve_imm_selective_upto_8")]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_wb = gen_reg_rtx (SImode);
+  emit_insn (gen_mve_vdwdupq_wb_u<mode>_insn (operands[0], ignore_wb,
+                                             operands[1], operands[2],
+                                             operands[3]));
+  DONE;
+})
+
+;;
+;; [vdwdupq_wb_u])
+;;
+(define_expand "mve_vdwdupq_wb_u<mode>"
+ [(match_operand:SI 0 "s_register_operand")
+  (match_operand:SI 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "mve_imm_selective_upto_8")
+  (unspec:MVE_2 [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_vec = gen_reg_rtx (<MODE>mode);
+  emit_insn (gen_mve_vdwdupq_wb_u<mode>_insn (ignore_vec, operands[0],
+                                             operands[1], operands[2],
+                                             operands[3]));
+  DONE;
+})
+
+;;
+;; [vdwdupq_wb_u_insn])
+;;
+(define_insn "mve_vdwdupq_wb_u<mode>_insn"
+  [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:SI 2 "s_register_operand" "1")
+                      (match_operand:SI 3 "s_register_operand" "r")
+                      (match_operand:SI 4 "mve_imm_selective_upto_8" "Rg")]
+        VDWDUPQ))
+   (set (match_operand:SI 1 "s_register_operand" "=e")
+       (unspec:SI [(match_dup 2)
+                   (match_dup 3)
+                   (match_dup 4)]
+        VDWDUPQ))]
+  "TARGET_HAVE_MVE"
+  "vdwdup.u%#<V_sz_elem>\t%q0, %2, %3, %4"
+)
+
+;;
+;; [vdwdupq_m_n_u])
+;;
+(define_expand "mve_vdwdupq_m_n_u<mode>"
+ [(match_operand:MVE_2 0 "s_register_operand")
+  (match_operand:MVE_2 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "s_register_operand")
+  (match_operand:SI 4 "mve_imm_selective_upto_8")
+  (match_operand:HI 5 "vpr_register_operand")]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_wb = gen_reg_rtx (SImode);
+  emit_insn (gen_mve_vdwdupq_m_wb_u<mode>_insn (operands[0], ignore_wb,
+                                               operands[1], operands[2],
+                                               operands[3], operands[4],
+                                               operands[5]));
+  DONE;
+})
+
+;;
+;; [vdwdupq_m_wb_u])
+;;
+(define_expand "mve_vdwdupq_m_wb_u<mode>"
+ [(match_operand:SI 0 "s_register_operand")
+  (match_operand:MVE_2 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "s_register_operand")
+  (match_operand:SI 4 "mve_imm_selective_upto_8")
+  (match_operand:HI 5 "vpr_register_operand")]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_vec = gen_reg_rtx (<MODE>mode);
+  emit_insn (gen_mve_vdwdupq_m_wb_u<mode>_insn (ignore_vec, operands[0],
+                                               operands[1], operands[2],
+                                               operands[3], operands[4],
+                                               operands[5]));
+  DONE;
+})
+
+;;
+;; [vdwdupq_m_wb_u_insn])
+;;
+(define_insn "mve_vdwdupq_m_wb_u<mode>_insn"
+  [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:MVE_2 2 "s_register_operand" "w")
+                      (match_operand:SI 3 "s_register_operand" "1")
+                      (match_operand:SI 4 "s_register_operand" "r")
+                      (match_operand:SI 5 "mve_imm_selective_upto_8" "Rg")
+                      (match_operand:HI 6 "vpr_register_operand" "Up")]
+        VDWDUPQ_M))
+   (set (match_operand:SI 1 "s_register_operand" "=e")
+       (unspec:SI [(match_dup 2)
+                   (match_dup 3)
+                   (match_dup 4)
+                   (match_dup 5)
+                   (match_dup 6)]
+        VDWDUPQ_M))
+  ]
+  "TARGET_HAVE_MVE"
+  "vpst\;\tvdwdupt.u%#<V_sz_elem>\t%q2, %3, %4, %5"
+  [(set_attr "type" "mve_move")
+   (set_attr "length""8")])
+
+;;
+;; [viwdupq_n_u])
+;;
+(define_expand "mve_viwdupq_n_u<mode>"
+ [(match_operand:MVE_2 0 "s_register_operand")
+  (match_operand:SI 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "mve_imm_selective_upto_8")]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_wb = gen_reg_rtx (SImode);
+  emit_insn (gen_mve_viwdupq_wb_u<mode>_insn (operands[0], ignore_wb,
+                                             operands[1], operands[2],
+                                             operands[3]));
+  DONE;
+})
+
+;;
+;; [viwdupq_wb_u])
+;;
+(define_expand "mve_viwdupq_wb_u<mode>"
+ [(match_operand:SI 0 "s_register_operand")
+  (match_operand:SI 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "mve_imm_selective_upto_8")
+  (unspec:MVE_2 [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_vec = gen_reg_rtx (<MODE>mode);
+  emit_insn (gen_mve_viwdupq_wb_u<mode>_insn (ignore_vec, operands[0],
+                                             operands[1], operands[2],
+                                             operands[3]));
+  DONE;
+})
+
+;;
+;; [viwdupq_wb_u_insn])
+;;
+(define_insn "mve_viwdupq_wb_u<mode>_insn"
+  [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:SI 2 "s_register_operand" "1")
+                      (match_operand:SI 3 "s_register_operand" "r")
+                      (match_operand:SI 4 "mve_imm_selective_upto_8" "Rg")]
+        VIWDUPQ))
+   (set (match_operand:SI 1 "s_register_operand" "=e")
+       (unspec:SI [(match_dup 2)
+                   (match_dup 3)
+                   (match_dup 4)]
+        VIWDUPQ))]
+  "TARGET_HAVE_MVE"
+  "viwdup.u%#<V_sz_elem>\t%q0, %2, %3, %4"
+)
+
+;;
+;; [viwdupq_m_n_u])
+;;
+(define_expand "mve_viwdupq_m_n_u<mode>"
+ [(match_operand:MVE_2 0 "s_register_operand")
+  (match_operand:MVE_2 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "s_register_operand")
+  (match_operand:SI 4 "mve_imm_selective_upto_8")
+  (match_operand:HI 5 "vpr_register_operand")]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_wb = gen_reg_rtx (SImode);
+  emit_insn (gen_mve_viwdupq_m_wb_u<mode>_insn (operands[0], ignore_wb,
+                                               operands[1], operands[2],
+                                               operands[3], operands[4],
+                                               operands[5]));
+  DONE;
+})
+
+;;
+;; [viwdupq_m_wb_u])
+;;
+(define_expand "mve_viwdupq_m_wb_u<mode>"
+ [(match_operand:SI 0 "s_register_operand")
+  (match_operand:MVE_2 1 "s_register_operand")
+  (match_operand:SI 2 "s_register_operand")
+  (match_operand:SI 3 "s_register_operand")
+  (match_operand:SI 4 "mve_imm_selective_upto_8")
+  (match_operand:HI 5 "vpr_register_operand")]
+ "TARGET_HAVE_MVE"
+{
+  rtx ignore_vec = gen_reg_rtx (<MODE>mode);
+  emit_insn (gen_mve_viwdupq_m_wb_u<mode>_insn (ignore_vec, operands[0],
+                                               operands[1], operands[2],
+                                               operands[3], operands[4],
+                                               operands[5]));
+  DONE;
+})
+
+;;
+;; [viwdupq_m_wb_u_insn])
+;;
+(define_insn "mve_viwdupq_m_wb_u<mode>_insn"
+  [(set (match_operand:MVE_2 0 "s_register_operand" "=w")
+       (unspec:MVE_2 [(match_operand:MVE_2 2 "s_register_operand" "w")
+                      (match_operand:SI 3 "s_register_operand" "1")
+                      (match_operand:SI 4 "s_register_operand" "r")
+                      (match_operand:SI 5 "mve_imm_selective_upto_8" "Rg")
+                      (match_operand:HI 6 "vpr_register_operand" "Up")]
+        VIWDUPQ_M))
+   (set (match_operand:SI 1 "s_register_operand" "=e")
+       (unspec:SI [(match_dup 2)
+                   (match_dup 3)
+                   (match_dup 4)
+                   (match_dup 5)
+                   (match_dup 6)]
+        VIWDUPQ_M))
+  ]
+  "TARGET_HAVE_MVE"
+  "vpst\;\tviwdupt.u%#<V_sz_elem>\t%q2, %3, %4, %5"
+  [(set_attr "type" "mve_move")
+   (set_attr "length""8")])
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog

index 52008db..4506849 100644 (file)
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,4 +1,57 @@
  2020-03-20  Srinath Parvathaneni  <srinath.parvathaneni@arm.com>
+            Andre Vieira  <andre.simoesdiasvieira@arm.com>
+            Mihail Ionescu  <mihail.ionescu@arm.com>
+
+       * gcc.target/arm/mve/intrinsics/vddupq_m_n_u16.c: New test.
+       * gcc.target/arm/mve/intrinsics/vddupq_m_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_m_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_m_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_m_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_m_wb_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_n_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vddupq_wb_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_n_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vdwdupq_wb_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_m_n_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_m_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_m_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_m_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_m_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_m_wb_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_n_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/vidupq_wb_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_m_n_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_m_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_m_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_n_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_n_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_n_u8.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_wb_u16.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_wb_u32.c: Likewise.
+       * gcc.target/arm/mve/intrinsics/viwdupq_wb_u8.c: Likewise.
+
+2020-03-20  Srinath Parvathaneni  <srinath.parvathaneni@arm.com>
  
         * gcc.target/arm/mve/intrinsics/vuninitializedq_float.c: New test.
         * gcc.target/arm/mve/intrinsics/vuninitializedq_float1.c: Likewise.
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u16.c

new file mode 100644 (file)

index 0000000..ba875c6
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vddupq_m_n_u16 (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vddupq_m (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u32.c

new file mode 100644 (file)

index 0000000..618a5e2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vddupq_m_n_u32 (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vddupq_m (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u8.c

new file mode 100644 (file)

index 0000000..2ee01c3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vddupq_m_n_u8 (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vddupq_m (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u16.c

new file mode 100644 (file)

index 0000000..08f6a55
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vddupq_m_wb_u16 (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vddupq_m (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u32.c

new file mode 100644 (file)

index 0000000..16b471e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vddupq_m_wb_u32 (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vddupq_m (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u8.c

new file mode 100644 (file)

index 0000000..ade9fb8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vddupq_m_wb_u8 (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vddupq_m (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vddupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u16.c

new file mode 100644 (file)

index 0000000..b3d44f5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t a)
+{
+  return vddupq_n_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vddup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t a)
+{
+  return vddupq_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vddup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u32.c

new file mode 100644 (file)

index 0000000..163fc7f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t a)
+{
+  return vddupq_n_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t a)
+{
+  return vddupq_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u8.c

new file mode 100644 (file)

index 0000000..8309648
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t a)
+{
+  return vddupq_n_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t a)
+{
+  return vddupq_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u16.c

new file mode 100644 (file)

index 0000000..469daad
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t *a)
+{
+  return vddupq_wb_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vddup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t *a)
+{
+  return vddupq_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vddup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u32.c

new file mode 100644 (file)

index 0000000..69f365b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t *a)
+{
+  return vddupq_wb_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t *a)
+{
+  return vddupq_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u8.c

new file mode 100644 (file)

index 0000000..8d7ceb5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t *a)
+{
+  return vddupq_wb_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t *a)
+{
+  return vddupq_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vddup.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u16.c

new file mode 100644 (file)

index 0000000..4479065
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u32.c

new file mode 100644 (file)

index 0000000..874e864
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u8.c

new file mode 100644 (file)

index 0000000..7cb0780
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u16.c

new file mode 100644 (file)

index 0000000..bc8885f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t * a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 8, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t * a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 8, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u32.c

new file mode 100644 (file)

index 0000000..50ed948
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t * a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t * a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u8.c

new file mode 100644 (file)

index 0000000..839280e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t * a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 2, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t * a, uint32_t b, mve_pred16_t p)
+{
+  return vdwdupq_m (inactive, a, b, 2, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vdwdupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u16.c

new file mode 100644 (file)

index 0000000..dfb75ed
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t a, uint32_t b)
+{
+  return vdwdupq_n_u16 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t a, uint32_t b)
+{
+  return vdwdupq_u16 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u32.c

new file mode 100644 (file)

index 0000000..2597dbd
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t a, uint32_t b)
+{
+  return vdwdupq_n_u32 (a, b, 8);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t a, uint32_t b)
+{
+  return vdwdupq_u32 (a, b, 8);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u8.c

new file mode 100644 (file)

index 0000000..4a4bcb2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t a, uint32_t b)
+{
+  return vdwdupq_n_u8 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t a, uint32_t b)
+{
+  return vdwdupq_u8 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u16.c

new file mode 100644 (file)

index 0000000..9c4506a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t *a, uint32_t b)
+{
+  return vdwdupq_wb_u16 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t *a, uint32_t b)
+{
+  return vdwdupq_u16 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u32.c

new file mode 100644 (file)

index 0000000..782a6f4
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t *a, uint32_t b)
+{
+  return vdwdupq_wb_u32 (a, b, 8);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t *a, uint32_t b)
+{
+  return vdwdupq_u32 (a, b, 8);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u8.c

new file mode 100644 (file)

index 0000000..6a4e428
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t *a, uint32_t b)
+{
+  return vdwdupq_wb_u8 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t *a, uint32_t b)
+{
+  return vdwdupq_u8 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "vdwdup.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u16.c

new file mode 100644 (file)

index 0000000..60449de
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vidupq_m_n_u16 (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vidupq_m (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u32.c

new file mode 100644 (file)

index 0000000..1d358f9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vidupq_m_n_u32 (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vidupq_m (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u8.c

new file mode 100644 (file)

index 0000000..d32b8c5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vidupq_m_n_u8 (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t a, mve_pred16_t p)
+{
+  return vidupq_m (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u16.c

new file mode 100644 (file)

index 0000000..0b34b15
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vidupq_m_wb_u16 (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vidupq_m (inactive, a, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u32.c

new file mode 100644 (file)

index 0000000..cc6d6a1
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vidupq_m_wb_u32 (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vidupq_m (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u8.c

new file mode 100644 (file)

index 0000000..d6ed263
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vidupq_m_wb_u8 (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t *a, mve_pred16_t p)
+{
+  return vidupq_m (inactive, a, 1, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "vidupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u16.c

new file mode 100644 (file)

index 0000000..2bfecbb
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t a)
+{
+  return vidupq_n_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vidup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t a)
+{
+  return vidupq_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vidup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u32.c

new file mode 100644 (file)

index 0000000..f93aab5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t a)
+{
+  return vidupq_n_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t a)
+{
+  return vidupq_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u8.c

new file mode 100644 (file)

index 0000000..397d5e5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t a)
+{
+  return vidupq_n_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t a)
+{
+  return vidupq_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u16.c

new file mode 100644 (file)

index 0000000..d20b54d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t *a)
+{
+  return vidupq_wb_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vidup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t *a)
+{
+  return vidupq_u16 (a, 4);
+}
+
+/* { dg-final { scan-assembler "vidup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u32.c

new file mode 100644 (file)

index 0000000..c751a7c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t *a)
+{
+  return vidupq_wb_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t *a)
+{
+  return vidupq_u32 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u8.c

new file mode 100644 (file)

index 0000000..89c6da2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t *a)
+{
+  return vidupq_wb_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t *a)
+{
+  return vidupq_u8 (a, 1);
+}
+
+/* { dg-final { scan-assembler "vidup.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u16.c

new file mode 100644 (file)

index 0000000..b0317fa
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m_n_u16 (inactive, a, b, 2, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m (inactive, a, b, 2, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u32.c

new file mode 100644 (file)

index 0000000..2e0e0ad
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m_n_u32 (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u8.c

new file mode 100644 (file)

index 0000000..1ce1fff
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m_n_u8 (inactive, a, b, 8, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m (inactive, a, b, 8, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u16.c

new file mode 100644 (file)

index 0000000..834d3a9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u16.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t inactive, uint32_t *a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m_wb_u16 (inactive, a, b, 2, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u16"  }  } */
+
+uint16x8_t
+foo1 (uint16x8_t inactive, uint32_t *a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m (inactive, a, b, 2, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u32.c

new file mode 100644 (file)

index 0000000..75fcd9a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u32.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t inactive, uint32_t *a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m_wb_u32 (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32x4_t inactive, uint32_t *a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m (inactive, a, b, 4, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u8.c

new file mode 100644 (file)

index 0000000..a5ea588
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u8.c
@@ -0,0 +1,24 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t inactive, uint32_t *a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m_wb_u8 (inactive, a, b, 8, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u8"  }  } */
+
+uint8x16_t
+foo1 (uint8x16_t inactive, uint32_t *a, uint32_t b, mve_pred16_t p)
+{
+  return viwdupq_m (inactive, a, b, 8, p);
+}
+
+/* { dg-final { scan-assembler "vpst" } } */
+/* { dg-final { scan-assembler "viwdupt.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u16.c

new file mode 100644 (file)

index 0000000..52536d3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t a, uint32_t b)
+{
+  return viwdupq_n_u16 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "viwdup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t a, uint32_t b)
+{
+  return viwdupq_u16 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "viwdup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u32.c

new file mode 100644 (file)

index 0000000..49b15be
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t a, uint32_t b)
+{
+  return viwdupq_n_u32 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "viwdup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t a, uint32_t b)
+{
+  return viwdupq_u32 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "viwdup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u8.c

new file mode 100644 (file)

index 0000000..4beff7a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t a, uint32_t b)
+{
+  return viwdupq_n_u8 (a, b, 1);
+}
+
+/* { dg-final { scan-assembler "viwdup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t a, uint32_t b)
+{
+  return viwdupq_u8 (a, b, 1);
+}
+
+/* { dg-final { scan-assembler "viwdup.u8"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u16.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u16.c

new file mode 100644 (file)

index 0000000..0a88261
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u16.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint32_t * a, uint32_t b)
+{
+  return viwdupq_wb_u16 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "viwdup.u16"  }  } */
+
+uint16x8_t
+foo1 (uint32_t * a, uint32_t b)
+{
+  return viwdupq_u16 (a, b, 4);
+}
+
+/* { dg-final { scan-assembler "viwdup.u16"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u32.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u32.c

new file mode 100644 (file)

index 0000000..37e4f34
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u32.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32_t * a, uint32_t b)
+{
+  return viwdupq_wb_u32 (a, b, 8);
+}
+
+/* { dg-final { scan-assembler "viwdup.u32"  }  } */
+
+uint32x4_t
+foo1 (uint32_t * a, uint32_t b)
+{
+  return viwdupq_u32 (a, b, 8);
+}
+
+/* { dg-final { scan-assembler "viwdup.u32"  }  } */
diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u8.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u8.c

new file mode 100644 (file)

index 0000000..810bff9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u8.c
@@ -0,0 +1,22 @@
+/* { dg-do compile  } */
+/* { dg-require-effective-target arm_v8_1m_mve_ok } */
+/* { dg-add-options arm_v8_1m_mve } */
+/* { dg-additional-options "-O2" } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint32_t * a, uint32_t b)
+{
+  return viwdupq_wb_u8 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "viwdup.u8"  }  } */
+
+uint8x16_t
+foo1 (uint32_t * a, uint32_t b)
+{
+  return viwdupq_u8 (a, b, 2);
+}
+
+/* { dg-final { scan-assembler "viwdup.u8"  }  } */
author	Srinath Parvathaneni <srinath.parvathaneni@arm.com>
	Fri, 20 Mar 2020 11:58:30 +0000 (11:58 +0000)
committer	Kyrylo Tkachov <kyrylo.tkachov@arm.com>
	Fri, 20 Mar 2020 11:58:30 +0000 (11:58 +0000)
gcc/ChangeLog		patch \| blob \| history
gcc/config/arm/arm-builtins.c		patch \| blob \| history
gcc/config/arm/arm_mve.h		patch \| blob \| history
gcc/config/arm/arm_mve_builtins.def		patch \| blob \| history
gcc/config/arm/mve.md		patch \| blob \| history
gcc/testsuite/ChangeLog		patch \| blob \| history
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_m_wb_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vddupq_wb_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_m_wb_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vdwdupq_wb_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_m_wb_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/vidupq_wb_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_m_wb_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_n_u8.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u16.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u32.c	[new file with mode: 0644]	patch \| blob
gcc/testsuite/gcc.target/arm/mve/intrinsics/viwdupq_wb_u8.c	[new file with mode: 0644]	patch \| blob