Re: [PATCH][AArch64] Implement ACLE Data Intrinsics

[Richard Sandiford <richard.sandiford@arm.com>]

"Andre Vieira (lists)" <andre.simoesdiasvieira@arm.com> writes:
> On 29/06/2022 08:18, Richard Sandiford wrote:
>>> +      break;
>>> +    case AARCH64_RBIT:
>>> +    case AARCH64_RBITL:
>>> +    case AARCH64_RBITLL:
>>> +      if (mode == SImode)
>>> +	icode = CODE_FOR_aarch64_rbitsi;
>>> +      else
>>> +	icode = CODE_FOR_aarch64_rbitdi;
>>> +      break;
>>> +    default:
>>> +      gcc_unreachable ();
>>> +    }
>>> +  expand_insn (icode, 2, ops);
>>> +  return target;
>> This needs to return ops[0].value instead, since "target" just suggests
>> a possible location.
>>
>> Could you add tests for a memory source and memory destination, e.g.:
>>
>> void test_clz_mem (uint32_t *a)
>> {
>>    *a = __clz (*a);
>> }
>>
>> Without tests like that, these comments probably just sound like a paper
>> exercise, but they should make a difference for memory sources (previous
>> review) and memory destinations (this round).
> I had locally tested it (with rev though because clz doesn't use that 
> code) and strangely it does seem to work for the memory destinations, 
> but that's just a simple test.
> It could very well go wrong with some more complex codegen, so I'll just 
> take your word and use ops[0].value.
>
> And yeah I didn't add the tests at the time, don't really know why, I'll 
> chuck it down to laziness :P
>>
>>> diff --git a/gcc/config/aarch64/arm_acle.h b/gcc/config/aarch64/arm_acle.h
>>> index 9775a48c65825b424d3eb442384f5ab87b734fd7..a044bc74553fcf2a49b71290083f3f072fd5a2ce 100644
>>> --- a/gcc/config/aarch64/arm_acle.h
>>> +++ b/gcc/config/aarch64/arm_acle.h
>>> @@ -28,6 +28,7 @@
>>>   #define _GCC_ARM_ACLE_H
>>>   
>>>   #include <stdint.h>
>>> +#include <stddef.h>
>>>   
>>>   #pragma GCC aarch64 "arm_acle.h"
>>>   
>>> @@ -35,6 +36,58 @@
>>>   extern "C" {
>>>   #endif
>>>   
>>> +#define _GCC_ARM_ACLE_ROR_FN(NAME, TYPE)				  \
>>> +__extension__ extern __inline TYPE					  \
>>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))	  \
>>> +NAME (TYPE __value, uint32_t __rotate)					  \
>>> +{									  \
>>> +  size_t __size = sizeof (TYPE) * __CHAR_BIT__;				  \
>>> +  __rotate = __rotate % __size;						  \
>>> +  return __value >> __rotate | __value << ((__size - __rotate) % __size); \
>>> +}
>>> +
>>> +_GCC_ARM_ACLE_ROR_FN (__ror, uint32_t)
>>> +_GCC_ARM_ACLE_ROR_FN (__rorl, unsigned long)
>>> +_GCC_ARM_ACLE_ROR_FN (__rorll, uint64_t)
>>> +
>>> +#undef _GCC_ARM_ACLE_ROR_FN
>>> +
>>> +#define _GCC_ARM_ACLE_DATA_FN(NAME, BUILTIN, ITYPE, RTYPE)	    \
>>> +__extension__ extern __inline RTYPE				    \
>>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
>>> +__##NAME (ITYPE __value)					    \
>>> +{								    \
>>> +  return __builtin_##BUILTIN (__value);				    \
>>> +}
>>> +
>>> +_GCC_ARM_ACLE_DATA_FN (clz, clz, uint32_t, unsigned int)
>>> +_GCC_ARM_ACLE_DATA_FN (clzl, clzl, unsigned long, unsigned int)
>>> +_GCC_ARM_ACLE_DATA_FN (clzll, clzll, uint64_t, unsigned int)
>>> +_GCC_ARM_ACLE_DATA_FN (cls, clrsb, uint32_t, unsigned int)
>>> +_GCC_ARM_ACLE_DATA_FN (clsl, clrsbl, unsigned long, unsigned int)
>>> +_GCC_ARM_ACLE_DATA_FN (clsll, clrsbll, uint64_t, unsigned int)
>>> +_GCC_ARM_ACLE_DATA_FN (rev16, aarch64_rev16, uint32_t, uint32_t)
>>> +_GCC_ARM_ACLE_DATA_FN (rev16l, aarch64_rev16l, unsigned long, unsigned long)
>>> +_GCC_ARM_ACLE_DATA_FN (rev16ll, aarch64_rev16ll, uint64_t, uint64_t)
>>> +_GCC_ARM_ACLE_DATA_FN (rbit, aarch64_rbit, uint32_t, uint32_t)
>>> +_GCC_ARM_ACLE_DATA_FN (rbitl, aarch64_rbitl, unsigned long, unsigned long)
>>> +_GCC_ARM_ACLE_DATA_FN (rbitll, aarch64_rbitll, uint64_t, uint64_t)
>>> +_GCC_ARM_ACLE_DATA_FN (revsh, bswap16, int16_t, uint16_t)
>> The return type should be int16_t.
> Nice catch!
>> The clz and cls tests have the old return types (same as the argument
>> types), but I guess that's a good thing, since it shows that we avoid
>> the redundant zero-extend in clzll and clsll.
> Yeah I noticed that too when I was adding the mem tests, but I did 
> change them though because at the time it just felt like an oversight, 
> though I too was pleasantly surprised GCC was managing to avoid the 
> zero-extending :)
> I then saw your comment and made me wonder whether I should keep the 
> wrong return types in... I haven't but happy to change them back if you 
> think it's a nice 'test' to have.

I thought it was OK/useful as it was, but I don't mind either way.

BTW, while trying it out locally, I noticed:

  aarch64_init_data_intrinsics

was called from the wrong place.  Since it's adding normal __builtin
functions, it should be called from aarch64_general_init_builtins
instead of handle_arm_acle_h.

handle_arm_acle_h is instead for cases where we want to simulate
C/C++ definitions of the ACLE intrinsics themselves (i.e. so that
the intrinsics themselves are built-in functions, rather than
inline wrappers around built-in functions).

OK with that change, thanks.

Thanks,
Richard

> Regards,
> Andre
>
> diff --git a/gcc/config/aarch64/aarch64-builtins.cc b/gcc/config/aarch64/aarch64-builtins.cc
> index e0a741ac663188713e21f457affa57217d074783..bb5d97c8fc6402635270df851a949cabeecaa5e8 100644
> --- a/gcc/config/aarch64/aarch64-builtins.cc
> +++ b/gcc/config/aarch64/aarch64-builtins.cc
> @@ -613,6 +613,12 @@ enum aarch64_builtins
>    AARCH64_LS64_BUILTIN_ST64B,
>    AARCH64_LS64_BUILTIN_ST64BV,
>    AARCH64_LS64_BUILTIN_ST64BV0,
> +  AARCH64_REV16,
> +  AARCH64_REV16L,
> +  AARCH64_REV16LL,
> +  AARCH64_RBIT,
> +  AARCH64_RBITL,
> +  AARCH64_RBITLL,
>    AARCH64_BUILTIN_MAX
>  };
>  
> @@ -1664,10 +1670,41 @@ aarch64_init_ls64_builtins (void)
>        = aarch64_general_add_builtin (data[i].name, data[i].type, data[i].code);
>  }
>  
> +static void
> +aarch64_init_data_intrinsics (void)
> +{
> +  tree uint32_fntype = build_function_type_list (uint32_type_node,
> +						 uint32_type_node, NULL_TREE);
> +  tree ulong_fntype = build_function_type_list (long_unsigned_type_node,
> +						long_unsigned_type_node,
> +						NULL_TREE);
> +  tree uint64_fntype = build_function_type_list (uint64_type_node,
> +						 uint64_type_node, NULL_TREE);
> +  aarch64_builtin_decls[AARCH64_REV16]
> +    = aarch64_general_add_builtin ("__builtin_aarch64_rev16", uint32_fntype,
> +				   AARCH64_REV16);
> +  aarch64_builtin_decls[AARCH64_REV16L]
> +    = aarch64_general_add_builtin ("__builtin_aarch64_rev16l", ulong_fntype,
> +				   AARCH64_REV16L);
> +  aarch64_builtin_decls[AARCH64_REV16LL]
> +    = aarch64_general_add_builtin ("__builtin_aarch64_rev16ll", uint64_fntype,
> +				   AARCH64_REV16LL);
> +  aarch64_builtin_decls[AARCH64_RBIT]
> +    = aarch64_general_add_builtin ("__builtin_aarch64_rbit", uint32_fntype,
> +				   AARCH64_RBIT);
> +  aarch64_builtin_decls[AARCH64_RBITL]
> +    = aarch64_general_add_builtin ("__builtin_aarch64_rbitl", ulong_fntype,
> +				   AARCH64_RBITL);
> +  aarch64_builtin_decls[AARCH64_RBITLL]
> +    = aarch64_general_add_builtin ("__builtin_aarch64_rbitll", uint64_fntype,
> +				   AARCH64_RBITLL);
> +}
> +
>  /* Implement #pragma GCC aarch64 "arm_acle.h".  */
>  void
>  handle_arm_acle_h (void)
>  {
> +  aarch64_init_data_intrinsics ();
>    if (TARGET_LS64)
>      aarch64_init_ls64_builtins ();
>  }
> @@ -2394,6 +2431,37 @@ aarch64_expand_builtin_memtag (int fcode, tree exp, rtx target)
>    return target;
>  }
>  
> +/* Function to expand an expression EXP which calls one of the ACLE Data
> +   Intrinsic builtins FCODE with the result going to TARGET.  */
> +static rtx
> +aarch64_expand_builtin_data_intrinsic (unsigned int fcode, tree exp, rtx target)
> +{
> +  expand_operand ops[2];
> +  machine_mode mode = GET_MODE (target);
> +  create_output_operand (&ops[0], target, mode);
> +  create_input_operand (&ops[1], expand_normal (CALL_EXPR_ARG (exp, 0)), mode);
> +  enum insn_code icode;
> +
> +  switch (fcode)
> +    {
> +    case AARCH64_REV16:
> +    case AARCH64_REV16L:
> +    case AARCH64_REV16LL:
> +      icode = code_for_aarch64_rev16 (mode);
> +      break;
> +    case AARCH64_RBIT:
> +    case AARCH64_RBITL:
> +    case AARCH64_RBITLL:
> +      icode = code_for_aarch64_rbit (mode);
> +      break;
> +    default:
> +      gcc_unreachable ();
> +    }
> +
> +  expand_insn (icode, 2, ops);
> +  return ops[0].value;
> +}
> +
>  /* Expand an expression EXP as fpsr or fpcr setter (depending on
>     UNSPEC) using MODE.  */
>  static void
> @@ -2551,6 +2619,9 @@ aarch64_general_expand_builtin (unsigned int fcode, tree exp, rtx target,
>    if (fcode >= AARCH64_MEMTAG_BUILTIN_START
>        && fcode <= AARCH64_MEMTAG_BUILTIN_END)
>      return aarch64_expand_builtin_memtag (fcode, exp, target);
> +  if (fcode >= AARCH64_REV16
> +      && fcode <= AARCH64_RBITLL)
> +    return aarch64_expand_builtin_data_intrinsic (fcode, exp, target);
>  
>    gcc_unreachable ();
>  }
> diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
> index acec8c1146765c0fac73c15351853324b8f03209..ef0aed25c6b26eff61f9f6030dc5921a534e3d19 100644
> --- a/gcc/config/aarch64/aarch64.md
> +++ b/gcc/config/aarch64/aarch64.md
> @@ -4950,7 +4950,7 @@ (define_expand "ffs<mode>2"
>      rtx ccreg = aarch64_gen_compare_reg (EQ, operands[1], const0_rtx);
>      rtx x = gen_rtx_NE (VOIDmode, ccreg, const0_rtx);
>  
> -    emit_insn (gen_rbit<mode>2 (operands[0], operands[1]));
> +    emit_insn (gen_aarch64_rbit (<MODE>mode, operands[0], operands[1]));
>      emit_insn (gen_clz<mode>2 (operands[0], operands[0]));
>      emit_insn (gen_csinc3<mode>_insn (operands[0], x, operands[0], const0_rtx));
>      DONE;
> @@ -4996,7 +4996,7 @@ (define_insn "clrsb<mode>2"
>    [(set_attr "type" "clz")]
>  )
>  
> -(define_insn "rbit<mode>2"
> +(define_insn "@aarch64_rbit<mode>"
>    [(set (match_operand:GPI 0 "register_operand" "=r")
>  	(unspec:GPI [(match_operand:GPI 1 "register_operand" "r")] UNSPEC_RBIT))]
>    ""
> @@ -5017,7 +5017,7 @@ (define_insn_and_split "ctz<mode>2"
>    "reload_completed"
>    [(const_int 0)]
>    "
> -  emit_insn (gen_rbit<mode>2 (operands[0], operands[1]));
> +  emit_insn (gen_aarch64_rbit (<MODE>mode, operands[0], operands[1]));
>    emit_insn (gen_clz<mode>2 (operands[0], operands[0]));
>    DONE;
>  ")
> @@ -6022,6 +6022,13 @@ (define_insn "bswaphi2"
>    [(set_attr "type" "rev")]
>  )
>  
> +(define_insn "@aarch64_rev16<mode>"
> +  [(set (match_operand:GPI 0 "register_operand" "=r")
> +	(unspec:GPI [(match_operand:GPI 1 "register_operand" "r")] UNSPEC_REV))]
> +  ""
> +  "rev16\\t%<w>0, %<w>1"
> +  [(set_attr "type" "rev")])
> +
>  (define_insn "*aarch64_bfxil<mode>"
>    [(set (match_operand:GPI 0 "register_operand" "=r,r")
>      (ior:GPI (and:GPI (match_operand:GPI 1 "register_operand" "r,0")
> diff --git a/gcc/config/aarch64/arm_acle.h b/gcc/config/aarch64/arm_acle.h
> index 9775a48c65825b424d3eb442384f5ab87b734fd7..d26e269cb843fe37ba789db09c40d06f53438cda 100644
> --- a/gcc/config/aarch64/arm_acle.h
> +++ b/gcc/config/aarch64/arm_acle.h
> @@ -28,6 +28,7 @@
>  #define _GCC_ARM_ACLE_H
>  
>  #include <stdint.h>
> +#include <stddef.h>
>  
>  #pragma GCC aarch64 "arm_acle.h"
>  
> @@ -35,6 +36,58 @@
>  extern "C" {
>  #endif
>  
> +#define _GCC_ARM_ACLE_ROR_FN(NAME, TYPE)				  \
> +__extension__ extern __inline TYPE					  \
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))	  \
> +NAME (TYPE __value, uint32_t __rotate)					  \
> +{									  \
> +  size_t __size = sizeof (TYPE) * __CHAR_BIT__;				  \
> +  __rotate = __rotate % __size;						  \
> +  return __value >> __rotate | __value << ((__size - __rotate) % __size); \
> +}
> +
> +_GCC_ARM_ACLE_ROR_FN (__ror, uint32_t)
> +_GCC_ARM_ACLE_ROR_FN (__rorl, unsigned long)
> +_GCC_ARM_ACLE_ROR_FN (__rorll, uint64_t)
> +
> +#undef _GCC_ARM_ACLE_ROR_FN
> +
> +#define _GCC_ARM_ACLE_DATA_FN(NAME, BUILTIN, ITYPE, RTYPE)	    \
> +__extension__ extern __inline RTYPE				    \
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
> +__##NAME (ITYPE __value)					    \
> +{								    \
> +  return __builtin_##BUILTIN (__value);				    \
> +}
> +
> +_GCC_ARM_ACLE_DATA_FN (clz, clz, uint32_t, unsigned int)
> +_GCC_ARM_ACLE_DATA_FN (clzl, clzl, unsigned long, unsigned int)
> +_GCC_ARM_ACLE_DATA_FN (clzll, clzll, uint64_t, unsigned int)
> +_GCC_ARM_ACLE_DATA_FN (cls, clrsb, uint32_t, unsigned int)
> +_GCC_ARM_ACLE_DATA_FN (clsl, clrsbl, unsigned long, unsigned int)
> +_GCC_ARM_ACLE_DATA_FN (clsll, clrsbll, uint64_t, unsigned int)
> +_GCC_ARM_ACLE_DATA_FN (rev16, aarch64_rev16, uint32_t, uint32_t)
> +_GCC_ARM_ACLE_DATA_FN (rev16l, aarch64_rev16l, unsigned long, unsigned long)
> +_GCC_ARM_ACLE_DATA_FN (rev16ll, aarch64_rev16ll, uint64_t, uint64_t)
> +_GCC_ARM_ACLE_DATA_FN (rbit, aarch64_rbit, uint32_t, uint32_t)
> +_GCC_ARM_ACLE_DATA_FN (rbitl, aarch64_rbitl, unsigned long, unsigned long)
> +_GCC_ARM_ACLE_DATA_FN (rbitll, aarch64_rbitll, uint64_t, uint64_t)
> +_GCC_ARM_ACLE_DATA_FN (revsh, bswap16, int16_t, int16_t)
> +_GCC_ARM_ACLE_DATA_FN (rev, bswap32, uint32_t, uint32_t)
> +_GCC_ARM_ACLE_DATA_FN (revll, bswap64, uint64_t, uint64_t)
> +
> +#undef _GCC_ARM_ACLE_DATA_FN
> +
> +__extension__ extern __inline unsigned long
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +__revl (unsigned long __value)
> +{
> +  if (sizeof (unsigned long) == 8)
> +    return __revll (__value);
> +  else
> +    return __rev (__value);
> +}
> +
>  #pragma GCC push_options
>  #pragma GCC target ("arch=armv8.3-a")
>  __extension__ extern __inline int32_t
> diff --git a/gcc/testsuite/gcc.target/aarch64/acle/data-intrinsics.c b/gcc/testsuite/gcc.target/aarch64/acle/data-intrinsics.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..e067ef20bbdc8993865b541aa99dccac6b03e6a0
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/acle/data-intrinsics.c
> @@ -0,0 +1,468 @@
> +/* Test the ACLE data intrinsics.  */
> +/* { dg-do assemble } */
> +/* { dg-additional-options "--save-temps -O1" } */
> +/* { dg-final { check-function-bodies "**" "" "" } } */
> +
> +#include "arm_acle.h"
> +
> +/*
> +** test_clz:
> +**	clz	w0, w0
> +**	ret
> +*/
> +
> +unsigned int test_clz (uint32_t a)
> +{
> +  return __clz (a);
> +}
> +
> +/*
> +** test_clzl:
> +**	clz	[wx]0, [wx]0
> +**	ret
> +*/
> +
> +unsigned int test_clzl (unsigned long a)
> +{
> +  return __clzl (a);
> +}
> +
> +/*
> +** test_clzll:
> +**	clz	x0, x0
> +**	ret
> +*/
> +
> +unsigned int test_clzll (uint64_t a)
> +{
> +  return __clzll (a);
> +}
> +
> +/*
> +** test_cls:
> +**	cls	w0, w0
> +**	ret
> +*/
> +
> +unsigned int test_cls (uint32_t a)
> +{
> +  return __cls (a);
> +}
> +
> +/*
> +** test_clsl:
> +**	cls	[wx]0, [wx]0
> +**	ret
> +*/
> +
> +unsigned int test_clsl (unsigned long a)
> +{
> +  return __clsl (a);
> +}
> +
> +/*
> +** test_clsll:
> +**	cls	x0, x0
> +**	ret
> +*/
> +
> +unsigned int test_clsll (uint64_t a)
> +{
> +  return __clsll (a);
> +}
> +
> +/*
> +** test_rbit:
> +**	rbit	w0, w0
> +**	ret
> +*/
> +
> +uint32_t test_rbit (uint32_t a)
> +{
> +  return __rbit (a);
> +}
> +
> +/*
> +** test_rbitl:
> +**	rbit	[wx]0, [wx]0
> +**	ret
> +*/
> +
> +unsigned long test_rbitl (unsigned long a)
> +{
> +  return __rbitl (a);
> +}
> +
> +/*
> +** test_rbitll:
> +**	rbit	x0, x0
> +**	ret
> +*/
> +
> +uint64_t test_rbitll (uint64_t a)
> +{
> +  return __rbitll (a);
> +}
> +
> +/*
> +** test_rev:
> +**	rev	w0, w0
> +**	ret
> +*/
> +
> +uint32_t test_rev (uint32_t a)
> +{
> +  return __rev (a);
> +}
> +
> +/*
> +** test_revl:
> +**	rev	[wx]0, [wx]0
> +**	ret
> +*/
> +
> +unsigned long test_revl (unsigned long a)
> +{
> +  return __revl (a);
> +}
> +
> +/*
> +** test_revll:
> +**	rev	x0, x0
> +**	ret
> +*/
> +
> +uint64_t test_revll (uint64_t a)
> +{
> +  return __revll (a);
> +}
> +
> +/*
> +** test_rev16:
> +**	rev16	w0, w0
> +**	ret
> +*/
> +
> +uint32_t test_rev16 (uint32_t a)
> +{
> +  return __rev16 (a);
> +}
> +
> +/*
> +** test_rev16l:
> +**	rev16	[wx]0, [wx]0
> +**	ret
> +*/
> +
> +unsigned long test_rev16l (unsigned long a)
> +{
> +  return __rev16l (a);
> +}
> +
> +/*
> +** test_rev16ll:
> +**	rev16	x0, x0
> +**	ret
> +*/
> +
> +uint64_t test_rev16ll (uint64_t a)
> +{
> +  return __rev16ll (a);
> +}
> +
> +/*
> +** test_ror:
> +**	ror	w0, w0, w1
> +**	ret
> +*/
> +
> +uint32_t test_ror (uint32_t a, uint32_t r)
> +{
> +  return __ror (a, r);
> +}
> +
> +/*
> +** test_rorl:
> +**	ror	[wx]0, [wx]0, [wx]1
> +**	ret
> +*/
> +
> +unsigned long test_rorl (unsigned long a, uint32_t r)
> +{
> +  return __rorl (a, r);
> +}
> +
> +/*
> +** test_rorll:
> +**	ror	x0, x0, x1
> +**	ret
> +*/
> +
> +uint64_t test_rorll (uint64_t a, uint32_t r)
> +{
> +  return __rorll (a, r);
> +}
> +
> +/*
> +** test_revsh:
> +**	rev16	w0, w0
> +**	ret
> +*/
> +
> +int16_t test_revsh (int16_t a)
> +{
> +  return __revsh (a);
> +}
> +
> +uint32_t *g32;
> +unsigned long *gul;
> +uint64_t *g64;
> +unsigned int *gui;
> +int16_t *g16;
> +
> +/*
> +** test_clz_mem:
> +**	...
> +**	clz	w[0-9]+, w[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_clz_mem (uint32_t *a)
> +{
> +  *gui = __clz (*a);
> +}
> +
> +/*
> +** test_clzl_mem:
> +**	...
> +**	clz	[wx][0-9]+, [wx][0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_clzl_mem (unsigned long *a)
> +{
> +  *gui = __clzl (*a);
> +}
> +
> +/*
> +** test_clzll_mem:
> +**	...
> +**	clz	x[0-9]+, x[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_clzll_mem (uint64_t *a)
> +{
> +  *gui = __clzll (*a);
> +}
> +
> +/*
> +** test_cls_mem:
> +**	...
> +**	cls	w[0-9]+, w[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_cls_mem (uint32_t *a)
> +{
> +  *gui = __cls (*a);
> +}
> +
> +/*
> +** test_clsl_mem:
> +**	...
> +**	cls	[wx][0-9]+, [wx][0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_clsl_mem (unsigned long *a)
> +{
> +  *gui = __clsl (*a);
> +}
> +
> +/*
> +** test_clsll_mem:
> +**	...
> +**	cls	x[0-9]+, x[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_clsll_mem (uint64_t *a)
> +{
> +  *gui = __clsll (*a);
> +}
> +
> +/*
> +** test_rbit_mem:
> +**	...
> +**	rbit	w[0-9]+, w[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rbit_mem (uint32_t *a)
> +{
> +  *g32 = __rbit (*a);
> +}
> +
> +/*
> +** test_rbitl_mem:
> +**	...
> +**	rbit	[wx][0-9]+, [wx][0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rbitl_mem (unsigned long *a)
> +{
> +  *gul = __rbitl (*a);
> +}
> +
> +/*
> +** test_rbitll_mem:
> +**	...
> +**	rbit	x[0-9]+, x[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rbitll_mem (uint64_t *a)
> +{
> +  *g64 = __rbitll (*a);
> +}
> +
> +/*
> +** test_rev_mem:
> +**	...
> +**	rev	w[0-9]+, w[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rev_mem (uint32_t *a)
> +{
> +  *g32 = __rev (*a);
> +}
> +
> +/*
> +** test_revl_mem:
> +**	...
> +**	rev	[wx][0-9]+, [wx][0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_revl_mem (unsigned long *a)
> +{
> +  *gul = __revl (*a);
> +}
> +
> +/*
> +** test_revll_mem:
> +**	...
> +**	rev	x[0-9]+, x[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_revll_mem (uint64_t *a)
> +{
> +  *g64 = __revll (*a);
> +}
> +
> +/*
> +** test_rev16_mem:
> +**	...
> +**	rev16	w[0-9]+, w[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rev16_mem (uint32_t *a)
> +{
> +  *g32 = __rev16 (*a);
> +}
> +
> +/*
> +** test_rev16l_mem:
> +**	...
> +**	rev16	[wx][0-9]+, [wx][0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rev16l_mem (unsigned long *a)
> +{
> +  *gul = __rev16l (*a);
> +}
> +
> +/*
> +** test_rev16ll_mem:
> +**	...
> +**	rev16	x[0-9]+, x[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rev16ll_mem (uint64_t *a)
> +{
> +  *g64 = __rev16ll (*a);
> +}
> +
> +/*
> +** test_ror_mem:
> +**	...
> +**	ror	w[0-9]+, w[0-9]+, w[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_ror_mem (uint32_t *a, uint32_t *r)
> +{
> +  *g32 = __ror (*a, *r);
> +}
> +
> +/*
> +** test_rorl_mem:
> +**	...
> +**	ror	[wx][0-9]+, [wx][0-9]+, [wx][0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rorl_mem (unsigned long *a, uint32_t *r)
> +{
> +  *gul = __rorl (*a, *r);
> +}
> +
> +/*
> +** test_rorll_mem:
> +**	...
> +**	ror	x[0-9]+, x[0-9]+, x[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_rorll_mem (uint64_t *a, uint32_t *r)
> +{
> +  *g64 = __rorll (*a, *r);
> +}
> +
> +/*
> +** test_revsh_mem:
> +**	...
> +**	rev16	w[0-9]+, w[0-9]+
> +**	...
> +**	ret
> +*/
> +
> +void test_revsh_mem (int16_t *a)
> +{
> +  *g16 = __revsh (*a);
> +}