[ARM][PR66791] Replace builtins for vclt and vcgt intrinsics in arm_neon.h

[ARM][PR66791] Replace builtins for vclt and vcgt intrinsics in arm_neon.h

[Prathamesh Kulkarni]

[-- Attachment #1: Type: text/plain, Size: 196 bytes --]

Hi,
This patch replaces calls to __builtin_neon_vcgt* by < and > in vclt
and vcgt intrinsics respectively, and removes entry to vcgt from
arm_neon_builtins.def.
OK to commit ?

Thanks,
Prathamesh

[-- Attachment #2: vclt-2.diff --]
[-- Type: application/octet-stream, Size: 8243 bytes --]

diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index 9569e1a4c9c..2980d9257af 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -2667,196 +2667,196 @@ __extension__ extern __inline uint8x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_s8 (int8x8_t __a, int8x8_t __b)
 {
-  return (uint8x8_t)__builtin_neon_vcgtv8qi (__a, __b);
+  return (uint8x8_t) (__a > __b);
 }
 
 __extension__ extern __inline uint16x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_s16 (int16x4_t __a, int16x4_t __b)
 {
-  return (uint16x4_t)__builtin_neon_vcgtv4hi (__a, __b);
+  return (uint16x4_t) (__a > __b);
 }
 
 __extension__ extern __inline uint32x2_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_s32 (int32x2_t __a, int32x2_t __b)
 {
-  return (uint32x2_t)__builtin_neon_vcgtv2si (__a, __b);
+  return (uint32x2_t) (__a > __b);
 }
 
 __extension__ extern __inline uint32x2_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_f32 (float32x2_t __a, float32x2_t __b)
 {
-  return (uint32x2_t)__builtin_neon_vcgtv2sf (__a, __b);
+  return (uint32x2_t) (__a > __b);
 }
 
 __extension__ extern __inline uint8x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_u8 (uint8x8_t __a, uint8x8_t __b)
 {
-  return (uint8x8_t)__builtin_neon_vcgtuv8qi ((int8x8_t) __a, (int8x8_t) __b);
+  return (__a > __b);
 }
 
 __extension__ extern __inline uint16x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_u16 (uint16x4_t __a, uint16x4_t __b)
 {
-  return (uint16x4_t)__builtin_neon_vcgtuv4hi ((int16x4_t) __a, (int16x4_t) __b);
+  return (__a > __b);
 }
 
 __extension__ extern __inline uint32x2_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_u32 (uint32x2_t __a, uint32x2_t __b)
 {
-  return (uint32x2_t)__builtin_neon_vcgtuv2si ((int32x2_t) __a, (int32x2_t) __b);
+  return (__a > __b);
 }
 
 __extension__ extern __inline uint8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_s8 (int8x16_t __a, int8x16_t __b)
 {
-  return (uint8x16_t)__builtin_neon_vcgtv16qi (__a, __b);
+  return (uint8x16_t) (__a > __b);
 }
 
 __extension__ extern __inline uint16x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_s16 (int16x8_t __a, int16x8_t __b)
 {
-  return (uint16x8_t)__builtin_neon_vcgtv8hi (__a, __b);
+  return (uint16x8_t) (__a > __b);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_s32 (int32x4_t __a, int32x4_t __b)
 {
-  return (uint32x4_t)__builtin_neon_vcgtv4si (__a, __b);
+  return (uint32x4_t) (__a > __b);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_f32 (float32x4_t __a, float32x4_t __b)
 {
-  return (uint32x4_t)__builtin_neon_vcgtv4sf (__a, __b);
+  return (uint32x4_t) (__a > __b);
 }
 
 __extension__ extern __inline uint8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
-  return (uint8x16_t)__builtin_neon_vcgtuv16qi ((int8x16_t) __a, (int8x16_t) __b);
+  return (__a > __b);
 }
 
 __extension__ extern __inline uint16x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
-  return (uint16x8_t)__builtin_neon_vcgtuv8hi ((int16x8_t) __a, (int16x8_t) __b);
+  return (__a > __b);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
-  return (uint32x4_t)__builtin_neon_vcgtuv4si ((int32x4_t) __a, (int32x4_t) __b);
+  return (__a > __b);
 }
 
 __extension__ extern __inline uint8x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_s8 (int8x8_t __a, int8x8_t __b)
 {
-  return (uint8x8_t)__builtin_neon_vcgtv8qi (__b, __a);
+  return (uint8x8_t) (__a < __b);
 }
 
 __extension__ extern __inline uint16x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_s16 (int16x4_t __a, int16x4_t __b)
 {
-  return (uint16x4_t)__builtin_neon_vcgtv4hi (__b, __a);
+  return (uint16x4_t) (__a < __b);
 }
 
 __extension__ extern __inline uint32x2_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_s32 (int32x2_t __a, int32x2_t __b)
 {
-  return (uint32x2_t)__builtin_neon_vcgtv2si (__b, __a);
+  return (uint32x2_t) (__a < __b);
 }
 
 __extension__ extern __inline uint32x2_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_f32 (float32x2_t __a, float32x2_t __b)
 {
-  return (uint32x2_t)__builtin_neon_vcgtv2sf (__b, __a);
+  return (uint32x2_t) (__a < __b);
 }
 
 __extension__ extern __inline uint8x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_u8 (uint8x8_t __a, uint8x8_t __b)
 {
-  return (uint8x8_t)__builtin_neon_vcgtuv8qi ((int8x8_t) __b, (int8x8_t) __a);
+  return (__a < __b);
 }
 
 __extension__ extern __inline uint16x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_u16 (uint16x4_t __a, uint16x4_t __b)
 {
-  return (uint16x4_t)__builtin_neon_vcgtuv4hi ((int16x4_t) __b, (int16x4_t) __a);
+  return (__a < __b);
 }
 
 __extension__ extern __inline uint32x2_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_u32 (uint32x2_t __a, uint32x2_t __b)
 {
-  return (uint32x2_t)__builtin_neon_vcgtuv2si ((int32x2_t) __b, (int32x2_t) __a);
+  return (__a < __b);
 }
 
 __extension__ extern __inline uint8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_s8 (int8x16_t __a, int8x16_t __b)
 {
-  return (uint8x16_t)__builtin_neon_vcgtv16qi (__b, __a);
+  return (uint8x16_t) (__a < __b);
 }
 
 __extension__ extern __inline uint16x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_s16 (int16x8_t __a, int16x8_t __b)
 {
-  return (uint16x8_t)__builtin_neon_vcgtv8hi (__b, __a);
+  return (uint16x8_t) (__a < __b);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_s32 (int32x4_t __a, int32x4_t __b)
 {
-  return (uint32x4_t)__builtin_neon_vcgtv4si (__b, __a);
+  return (uint32x4_t) (__a < __b);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_f32 (float32x4_t __a, float32x4_t __b)
 {
-  return (uint32x4_t)__builtin_neon_vcgtv4sf (__b, __a);
+  return (uint32x4_t) (__a < __b);
 }
 
 __extension__ extern __inline uint8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
-  return (uint8x16_t)__builtin_neon_vcgtuv16qi ((int8x16_t) __b, (int8x16_t) __a);
+  return (__a < __b);
 }
 
 __extension__ extern __inline uint16x8_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
-  return (uint16x8_t)__builtin_neon_vcgtuv8hi ((int16x8_t) __b, (int16x8_t) __a);
+  return (__a < __b);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
-  return (uint32x4_t)__builtin_neon_vcgtuv4si ((int32x4_t) __b, (int32x4_t) __a);
+  return (__a < __b);
 }
 
 __extension__ extern __inline uint32x2_t
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index 0ff0494b5da..5ec7f14b035 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -128,8 +128,6 @@ VAR3 (BINOP, vrsubhn, v8hi, v4si, v2di)
 VAR8 (BINOP, vceq, v8qi, v4hi, v2si, v2sf, v16qi, v8hi, v4si, v4sf)
 VAR8 (BINOP, vcge, v8qi, v4hi, v2si, v2sf, v16qi, v8hi, v4si, v4sf)
 VAR6 (BINOP, vcgeu, v8qi, v4hi, v2si, v16qi, v8hi, v4si)
-VAR8 (BINOP, vcgt, v8qi, v4hi, v2si, v2sf, v16qi, v8hi, v4si, v4sf)
-VAR6 (BINOP, vcgtu, v8qi, v4hi, v2si, v16qi, v8hi, v4si)
 VAR2 (BINOP, vcage, v2sf, v4sf)
 VAR2 (BINOP, vcagt, v2sf, v4sf)
 VAR2 (BINOP, vcage, v4hf, v8hf)

RE: [ARM][PR66791] Replace builtins for vclt and vcgt intrinsics in arm_neon.h

[Kyrylo Tkachov]

> -----Original Message-----
> From: Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
> Sent: 08 December 2020 11:04
> To: gcc Patches <gcc-patches@gcc.gnu.org>; Kyrylo Tkachov
> <Kyrylo.Tkachov@arm.com>
> Subject: [ARM][PR66791] Replace builtins for vclt and vcgt intrinsics in
> arm_neon.h
> 
> Hi,
> This patch replaces calls to __builtin_neon_vcgt* by < and > in vclt
> and vcgt intrinsics respectively, and removes entry to vcgt from
> arm_neon_builtins.def.
> OK to commit ?

Ok.
Thanks,
Kyrill

> 
> Thanks,
> Prathamesh