[PATCH] AArch64: Implement missing vcls intrinsics on unsigned types

[PATCH] AArch64: Implement missing vcls intrinsics on unsigned types

[Kyrylo Tkachov]

[-- Attachment #1: Type: text/plain, Size: 547 bytes --]

Hi all,

This patch implements some missing intrinsics that perform a CLS on unsigned SIMD types.

Bootstrapped and tested on aarch64-none-linux-gnu.

Committing to trunk and to active branches after testing.

Thanks,
Kyrill

2020-09-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

	PR target/71233
	* config/aarch64/arm_neon.h (vcls_u8, vcls_u16, vcls_u32,
	vclsq_u8, vclsq_u16, vclsq_u32): Define.

2020-09-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

	PR target/71233
	* gcc.target/aarch64/simd/vcls_unsigned_1.c: New test.

[-- Attachment #2: vcls-u.patch --]
[-- Type: application/octet-stream, Size: 2951 bytes --]

diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index e2c6a8c5da73851378afad410737e1ca272010ba..622852ca177e1122a521f0cb275ddb617554a462 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -14075,6 +14075,48 @@ vclsq_s32 (int32x4_t __a)
   return __builtin_aarch64_clrsbv4si (__a);
 }
 
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vcls_u8 (uint8x8_t __a)
+{
+  return __builtin_aarch64_clrsbv8qi ((int8x8_t) __a);
+}
+
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vcls_u16 (uint16x4_t __a)
+{
+  return __builtin_aarch64_clrsbv4hi ((int16x4_t) __a);
+}
+
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vcls_u32 (uint32x2_t __a)
+{
+  return __builtin_aarch64_clrsbv2si ((int32x2_t) __a);
+}
+
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vclsq_u8 (uint8x16_t __a)
+{
+  return __builtin_aarch64_clrsbv16qi ((int8x16_t) __a);
+}
+
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vclsq_u16 (uint16x8_t __a)
+{
+  return __builtin_aarch64_clrsbv8hi ((int16x8_t) __a);
+}
+
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vclsq_u32 (uint32x4_t __a)
+{
+  return __builtin_aarch64_clrsbv4si ((int32x4_t) __a);
+}
+
 /* vclz.  */
 
 __extension__ extern __inline int8x8_t
diff --git a/gcc/testsuite/gcc.target/aarch64/simd/vcls_unsigned_1.c b/gcc/testsuite/gcc.target/aarch64/simd/vcls_unsigned_1.c
new file mode 100644
index 0000000000000000000000000000000000000000..f7078d1a67c9dc5397af5c8a89cf629afdb11241
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/simd/vcls_unsigned_1.c
@@ -0,0 +1,54 @@
+/* { dg-do compile } */
+/* { dg-options "-O" } */
+
+#include <arm_neon.h>
+
+int16x8_t
+test_16x8 (uint16x8_t a)
+{
+  return vclsq_u16 (a);
+}
+
+/* { dg-final { scan-assembler-times "cls\\tv\[0-9\]+\.8h, v\[0-9\]+\.8h" 1 } } */
+
+
+int8x16_t
+test_8x16 (uint8x16_t a)
+{
+  return vclsq_u8 (a);
+}
+
+/* { dg-final { scan-assembler-times "cls\\tv\[0-9\]+\.16b, v\[0-9\]+\.16b" 1 } } */
+
+int32x4_t
+test_32x4 (uint32x4_t a)
+{
+  return vclsq_u32 (a);
+}
+
+/* { dg-final { scan-assembler-times "cls\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s" 1 } } */
+
+int16x4_t
+test_16x4 (uint16x4_t a)
+{
+  return vcls_u16 (a);
+}
+
+/* { dg-final { scan-assembler-times "cls\\tv\[0-9\]+\.4h, v\[0-9\]+\.4h" 1 } } */
+
+int8x8_t
+test_8x8 (uint8x8_t a)
+{
+  return vcls_u8 (a);
+}
+
+/* { dg-final { scan-assembler-times "cls\\tv\[0-9\]+\.8b, v\[0-9\]+\.8b" 1 } } */
+
+int32x2_t
+test32x2 (uint32x2_t a)
+{
+  return vcls_u32 (a);
+}
+
+/* { dg-final { scan-assembler-times "cls\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s" 1 } } */
+