diff mbox series

[v6,1/2] aarch64: Add AdvSIMD faminmax intrinsics

Message ID 20240829075150.4130637-2-saurabh.jha@arm.com
State New
Headers show
Series aarch64: Add support for AdvSIMD faminmax | expand

Commit Message

Saurabh Jha Aug. 29, 2024, 7:51 a.m. UTC
The AArch64 FEAT_FAMINMAX extension is optional from Armv9.2-a and
mandatory from Armv9.5-a. It introduces instructions for computing the
floating point absolute maximum and minimum of the two vectors element-wise.

This patch introduces AdvSIMD faminmax intrinsics. The intrinsics of
this extension are implemented as the following builtin functions:
* vamax_f16
* vamaxq_f16
* vamax_f32
* vamaxq_f32
* vamaxq_f64
* vamin_f16
* vaminq_f16
* vamin_f32
* vaminq_f32
* vaminq_f64

We are defining a new way to add AArch64 AdvSIMD intrinsics by listing
all the intrinsics in a .def file and then using that .def file to
initialise various data structures. This would lead to more concise code
and easier addition of the new AdvSIMD intrinsics in future.

The faminmax intrinsics are defined using the new approach

gcc/ChangeLog:

        * config/aarch64/aarch64-builtins.cc
	(ENTRY): Macro to parse the contents of
aarch64-simd-pragma-builtins.def.
	(enum aarch64_builtins): New enum values for faminmax builtins
via aarch64-simd-pragma-builtins.def.
        (aarch64_init_pragma_builtins): New function to define pragma builtins.
        (handle_arm_neon_h): Modify to call
aarch64_init_pragma_builtins.
        (aarch64_general_check_builtin_call): Modify to check whether
required flag is being used for pragma builtins.
        (aarch64_expand_pragma_builtin): New function to emit
instructions of pragma builtins.
        (aarch64_general_expand_builtin): Modify to call
aarch64_expand_pragma_builtin.
        * config/aarch64/aarch64-option-extensions.def
	(AARCH64_OPT_EXTENSION): Introduce new flag for this
extension.
        * config/aarch64/aarch64-simd.md
	(@aarch64_<faminmax_uns_op><mode>): Instruction pattern for
faminmax intrinsics.
        * config/aarch64/aarch64.h
	(TARGET_FAMINMAX): Introduce new flag for this extension.
        * config/aarch64/iterators.md: New iterators and unspecs.
        * config/arm/types.md: Introduce neon_fp_aminmax<q> attributes.
        * doc/invoke.texi: Document extension in AArch64 Options.
	* config/aarch64/aarch64-simd-pragma-builtins.def: New file to
	  list pragma builtins.

gcc/testsuite/ChangeLog:

        * gcc.target/aarch64/simd/faminmax-builtins-no-flag.c: New test.
        * gcc.target/aarch64/simd/faminmax-builtins.c: New test.
---
 gcc/config/aarch64/aarch64-builtins.cc        |  79 ++++++++++++
 .../aarch64/aarch64-option-extensions.def     |   2 +
 .../aarch64/aarch64-simd-pragma-builtins.def  |  31 +++++
 gcc/config/aarch64/aarch64-simd.md            |  11 ++
 gcc/config/aarch64/aarch64.h                  |   4 +
 gcc/config/aarch64/iterators.md               |   9 ++
 gcc/config/arm/types.md                       |   6 +
 gcc/doc/invoke.texi                           |   2 +
 .../aarch64/simd/faminmax-builtins-no-flag.c  |  10 ++
 .../aarch64/simd/faminmax-builtins.c          | 115 ++++++++++++++++++
 10 files changed, 269 insertions(+)
 create mode 100644 gcc/config/aarch64/aarch64-simd-pragma-builtins.def
 create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins-no-flag.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins.c

Comments

Kyrylo Tkachov Aug. 29, 2024, 9:10 a.m. UTC | #1
Hi Saurabh,

> On 29 Aug 2024, at 09:51, saurabh.jha@arm.com wrote:
> 
> External email: Use caution opening links or attachments
> 
> 
> The AArch64 FEAT_FAMINMAX extension is optional from Armv9.2-a and
> mandatory from Armv9.5-a. It introduces instructions for computing the
> floating point absolute maximum and minimum of the two vectors element-wise.
> 
> This patch introduces AdvSIMD faminmax intrinsics. The intrinsics of
> this extension are implemented as the following builtin functions:
> * vamax_f16
> * vamaxq_f16
> * vamax_f32
> * vamaxq_f32
> * vamaxq_f64
> * vamin_f16
> * vaminq_f16
> * vamin_f32
> * vaminq_f32
> * vaminq_f64
> 
> We are defining a new way to add AArch64 AdvSIMD intrinsics by listing
> all the intrinsics in a .def file and then using that .def file to
> initialise various data structures. This would lead to more concise code
> and easier addition of the new AdvSIMD intrinsics in future.
> 
> The faminmax intrinsics are defined using the new approach
> 
> gcc/ChangeLog:
> 
>        * config/aarch64/aarch64-builtins.cc
>        (ENTRY): Macro to parse the contents of
> aarch64-simd-pragma-builtins.def.
>        (enum aarch64_builtins): New enum values for faminmax builtins
> via aarch64-simd-pragma-builtins.def.
>        (aarch64_init_pragma_builtins): New function to define pragma builtins.
>        (handle_arm_neon_h): Modify to call
> aarch64_init_pragma_builtins.
>        (aarch64_general_check_builtin_call): Modify to check whether
> required flag is being used for pragma builtins.
>        (aarch64_expand_pragma_builtin): New function to emit
> instructions of pragma builtins.
>        (aarch64_general_expand_builtin): Modify to call
> aarch64_expand_pragma_builtin.
>        * config/aarch64/aarch64-option-extensions.def
>        (AARCH64_OPT_EXTENSION): Introduce new flag for this
> extension.
>        * config/aarch64/aarch64-simd.md
>        (@aarch64_<faminmax_uns_op><mode>): Instruction pattern for
> faminmax intrinsics.
>        * config/aarch64/aarch64.h
>        (TARGET_FAMINMAX): Introduce new flag for this extension.
>        * config/aarch64/iterators.md: New iterators and unspecs.
>        * config/arm/types.md: Introduce neon_fp_aminmax<q> attributes.
>        * doc/invoke.texi: Document extension in AArch64 Options.
>        * config/aarch64/aarch64-simd-pragma-builtins.def: New file to
>          list pragma builtins.
> 
> gcc/testsuite/ChangeLog:
> 
>        * gcc.target/aarch64/simd/faminmax-builtins-no-flag.c: New test.
>        * gcc.target/aarch64/simd/faminmax-builtins.c: New test.

Sorry for the back-and-forth, but I just realized, why can’t we reuse the existing aarch64_init_simd_intrinsics code in aarch64-builtins.cc <http://aarch64-builtins.cc/>?
It seems that it already handles most of the registration code, except that it doesn’t handle checking of arch extensions.
I think we should aim to refactor this a bit so that we can use that functionality.
But I appreciate that this would extend the scope of this patch a bit too much.
So I’m okay with this going in now, but it would be good to clean this area somewhat in a separate patch so that we can rely on just aarch64-simd-builtins.def, potentially augmented with extension information.
Thanks,
Kyrill


> ---
> gcc/config/aarch64/aarch64-builtins.cc        |  79 ++++++++++++
> .../aarch64/aarch64-option-extensions.def     |   2 +
> .../aarch64/aarch64-simd-pragma-builtins.def  |  31 +++++
> gcc/config/aarch64/aarch64-simd.md            |  11 ++
> gcc/config/aarch64/aarch64.h                  |   4 +
> gcc/config/aarch64/iterators.md               |   9 ++
> gcc/config/arm/types.md                       |   6 +
> gcc/doc/invoke.texi                           |   2 +
> .../aarch64/simd/faminmax-builtins-no-flag.c  |  10 ++
> .../aarch64/simd/faminmax-builtins.c          | 115 ++++++++++++++++++
> 10 files changed, 269 insertions(+)
> create mode 100644 gcc/config/aarch64/aarch64-simd-pragma-builtins.def
> create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins-no-flag.c
> create mode 100644 gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins.c
> 
> <v6-0001-aarch64-Add-AdvSIMD-faminmax-intrinsics.patch>
Richard Sandiford Aug. 29, 2024, 10:36 a.m. UTC | #2
Kyrylo Tkachov <ktkachov@nvidia.com> writes:
> Hi Saurabh,
>
>> On 29 Aug 2024, at 09:51, saurabh.jha@arm.com wrote:
>> 
>> External email: Use caution opening links or attachments
>> 
>> 
>> The AArch64 FEAT_FAMINMAX extension is optional from Armv9.2-a and
>> mandatory from Armv9.5-a. It introduces instructions for computing the
>> floating point absolute maximum and minimum of the two vectors element-wise.
>> 
>> This patch introduces AdvSIMD faminmax intrinsics. The intrinsics of
>> this extension are implemented as the following builtin functions:
>> * vamax_f16
>> * vamaxq_f16
>> * vamax_f32
>> * vamaxq_f32
>> * vamaxq_f64
>> * vamin_f16
>> * vaminq_f16
>> * vamin_f32
>> * vaminq_f32
>> * vaminq_f64
>> 
>> We are defining a new way to add AArch64 AdvSIMD intrinsics by listing
>> all the intrinsics in a .def file and then using that .def file to
>> initialise various data structures. This would lead to more concise code
>> and easier addition of the new AdvSIMD intrinsics in future.
>> 
>> The faminmax intrinsics are defined using the new approach
>> 
>> gcc/ChangeLog:
>> 
>>        * config/aarch64/aarch64-builtins.cc
>>        (ENTRY): Macro to parse the contents of
>> aarch64-simd-pragma-builtins.def.
>>        (enum aarch64_builtins): New enum values for faminmax builtins
>> via aarch64-simd-pragma-builtins.def.
>>        (aarch64_init_pragma_builtins): New function to define pragma builtins.
>>        (handle_arm_neon_h): Modify to call
>> aarch64_init_pragma_builtins.
>>        (aarch64_general_check_builtin_call): Modify to check whether
>> required flag is being used for pragma builtins.
>>        (aarch64_expand_pragma_builtin): New function to emit
>> instructions of pragma builtins.
>>        (aarch64_general_expand_builtin): Modify to call
>> aarch64_expand_pragma_builtin.
>>        * config/aarch64/aarch64-option-extensions.def
>>        (AARCH64_OPT_EXTENSION): Introduce new flag for this
>> extension.
>>        * config/aarch64/aarch64-simd.md
>>        (@aarch64_<faminmax_uns_op><mode>): Instruction pattern for
>> faminmax intrinsics.
>>        * config/aarch64/aarch64.h
>>        (TARGET_FAMINMAX): Introduce new flag for this extension.
>>        * config/aarch64/iterators.md: New iterators and unspecs.
>>        * config/arm/types.md: Introduce neon_fp_aminmax<q> attributes.
>>        * doc/invoke.texi: Document extension in AArch64 Options.
>>        * config/aarch64/aarch64-simd-pragma-builtins.def: New file to
>>          list pragma builtins.
>> 
>> gcc/testsuite/ChangeLog:
>> 
>>        * gcc.target/aarch64/simd/faminmax-builtins-no-flag.c: New test.
>>        * gcc.target/aarch64/simd/faminmax-builtins.c: New test.
>
> Sorry for the back-and-forth, but I just realized, why can’t we reuse the existing aarch64_init_simd_intrinsics code in aarch64-builtins.cc <http://aarch64-builtins.cc/>?
> It seems that it already handles most of the registration code, except that it doesn’t handle checking of arch extensions.
> I think we should aim to refactor this a bit so that we can use that functionality.
> But I appreciate that this would extend the scope of this patch a bit too much.
> So I’m okay with this going in now, but it would be good to clean this area somewhat in a separate patch so that we can rely on just aarch64-simd-builtins.def, potentially augmented with extension information.

Yeah, I agree we should consolidate this more.  I think the .def file
is a good long-term direction though, and it would be good to move the
existing builtins over to that.

As it stands, neither approach is a superset of the other.  The new
approach has the architecture requirements and unspec number, but the
old approach makes it easier to handle simplifications.  Neither approach
makes it particularly easy to encode the type signature.

So one of the tasks for the follow-on would be to find a good way of
expressing the type signature and a good way of handling things like
vget_high simplification (maybe via callbacks?).

On the patch itself:

>> diff --git a/gcc/config/aarch64/aarch64-builtins.cc b/gcc/config/aarch64/aarch64-builtins.cc
>> index eb878b933fe..61df394b881 100644
>> --- a/gcc/config/aarch64/aarch64-builtins.cc
>> +++ b/gcc/config/aarch64/aarch64-builtins.cc
>> @@ -757,6 +757,10 @@ typedef struct
>>  #define VAR1(T, N, MAP, FLAG, A) \
>>    AARCH64_SIMD_BUILTIN_##T##_##N##A,
>>  
>> +#undef ENTRY
>> +#define ENTRY(N, M, U, F) \
>> +  AARCH64_##N,
>> +
>>  enum aarch64_builtins
>>  {
>>    AARCH64_BUILTIN_MIN,
>> @@ -829,6 +833,10 @@ enum aarch64_builtins
>>    AARCH64_RBIT,
>>    AARCH64_RBITL,
>>    AARCH64_RBITLL,
>> +  /* Pragma builtins.  */
>> +  AARCH64_PRAGMA_BUILTIN_START,
>> +#include "aarch64-simd-pragma-builtins.def"
>> +  AARCH64_PRAGMA_BUILTIN_END,
>>    /* System register builtins.  */
>>    AARCH64_RSR,
>>    AARCH64_RSRP,
>> @@ -947,6 +955,7 @@ const char *aarch64_scalar_builtin_types[] = {
>>  
>>  extern GTY(()) aarch64_simd_type_info aarch64_simd_types[];
>>  
>> +#undef ENTRY
>>  #define ENTRY(E, M, Q, G)  \
>>    {E, "__" #E, #G "__" #E, NULL_TREE, NULL_TREE, E_##M##mode, qualifier_##Q},
>>  struct aarch64_simd_type_info aarch64_simd_types [] = {
>> @@ -1547,6 +1556,39 @@ aarch64_init_simd_builtin_functions (bool called_from_pragma)
>>      }
>>  }
>>  
>> +/* Initialize pragma builtins.  */
>> +
>> +typedef struct
>> +{
>> +  const char *name;
>> +  machine_mode mode;
>> +  int unspec;
>> +  aarch64_feature_flags required_extensions;
>> +} pragma_builtins_data;

The typedef struct { ... } thing is a holdover from C days.  New code
should just use structs directly.  The type name should have an "aarch64_"
prefix, to avoid accidental clashes with any future target-independent code.

So:

struct aarch64_pragma_builtins_data
{
  ...
};

>> +
>> +#undef ENTRY
>> +#define ENTRY(N, M, U, F) \
>> +  {#N, E_##M##mode, U, F},
>> +
>> +pragma_builtins_data pragma_builtins[] = {

Similarly, the variable name should have an "aarch64_" prefix.  It would
be good to make the array static.

>> +#include "aarch64-simd-pragma-builtins.def"
>> +};
>> +
>> +static void
>> +aarch64_init_pragma_builtins ()
>> +{
>> +  for (size_t i = 0; i < ARRAY_SIZE (pragma_builtins); ++i)
>> +    {
>> +      pragma_builtins_data data = pragma_builtins[i];
>> +      tree type = aarch64_simd_builtin_type (data.mode, qualifier_none);
>> +      tree fntype = build_function_type_list (type, type, type, NULL_TREE);
>> +      unsigned int code = AARCH64_PRAGMA_BUILTIN_START + i + 1;
>> +      const char *name = data.name;
>> +      aarch64_builtin_decls[code]
>> +	= aarch64_general_simulate_builtin (name, fntype, code);
>> +    }
>> +}
>> +
>>  /* Register the tuple type that contains NUM_VECTORS of the AdvSIMD type
>>     indexed by TYPE_INDEX.  */
>>  static void
>> @@ -1640,6 +1682,7 @@ handle_arm_neon_h (void)
>>  
>>    aarch64_init_simd_builtin_functions (true);
>>    aarch64_init_simd_intrinsics ();
>> +  aarch64_init_pragma_builtins ();
>>  }
>>  
>>  static void
>> @@ -2326,6 +2369,16 @@ aarch64_general_check_builtin_call (location_t location, vec<location_t>,
>>      return aarch64_check_required_extensions (location, decl,
>>  					      AARCH64_FL_MEMTAG);
>>  
>> +  if (code > AARCH64_PRAGMA_BUILTIN_START
>> +      && code < AARCH64_PRAGMA_BUILTIN_END)
>> +    {
>> +      unsigned int pragma_builtins_idx =
>> +	code - (AARCH64_PRAGMA_BUILTIN_START + 1);
>> +      aarch64_feature_flags flag =
>> +	pragma_builtins[pragma_builtins_idx].required_extensions;

It would be good to have a helper for this, say:

static const aarch64_pragma_builtins_data *
aarch64_get_pragma_builtin (int code)

that returns a pointer to the entry, or null if CODE isn't a pragma
builtin.  Then the code above could just be:

  if (auto *builtin = aarch64_get_pragma_builtin (int code))
    {
      auto flags = builtin->required_extensions;
      ...
    }

("flags" rather than "flag", since in general there can be multiple
requirements.)

>> +      return aarch64_check_required_extensions (location, decl, flag);
>> +    }
>> +
>>    return true;
>>  }
>>  
>> @@ -3189,6 +3242,29 @@ aarch64_expand_builtin_data_intrinsic (unsigned int fcode, tree exp, rtx target)
>>    return ops[0].value;
>>  }
>>  
>> +static rtx
>> +aarch64_expand_pragma_builtin (unsigned int fcode, tree exp, rtx target)

With the change above, this function could take the
const aarch64_pragma_builtins_data * (or const aarch64_pragma_builtins_data &)
as an extra argument.  Then...

>> +{
>> +  machine_mode mode = TYPE_MODE (TREE_TYPE (exp));
>> +  rtx op0 = force_reg (mode, expand_normal (CALL_EXPR_ARG (exp, 0)));
>> +  rtx op1 = force_reg (mode, expand_normal (CALL_EXPR_ARG (exp, 1)));
>> +
>> +  int unspec;
>> +  if (fcode >= AARCH64_vamax_f16 && fcode <= AARCH64_vamaxq_f64)
>> +    unspec = UNSPEC_FAMAX;
>> +  else if (fcode >= AARCH64_vamin_f16 && fcode <= AARCH64_vaminq_f64)
>> +    unspec = UNSPEC_FAMIN;
>> +  else
>> +    gcc_unreachable ();
>> +
>> +  enum insn_code icode = code_for_aarch64 (unspec, mode);

...we can get the code and mode from there.

>> +  rtx pat = GEN_FCN (icode) (target, op0, op1);

It probably doesn't matter for these builtins, but it's better to use
the expand_insn interface where possible, since it coerces the input and
output operands to the required form.  See existing uses in the same file
for examples.

Thanks for doing this!

Richard
diff mbox series

Patch

diff --git a/gcc/config/aarch64/aarch64-builtins.cc b/gcc/config/aarch64/aarch64-builtins.cc
index eb878b933fe..61df394b881 100644
--- a/gcc/config/aarch64/aarch64-builtins.cc
+++ b/gcc/config/aarch64/aarch64-builtins.cc
@@ -757,6 +757,10 @@  typedef struct
 #define VAR1(T, N, MAP, FLAG, A) \
   AARCH64_SIMD_BUILTIN_##T##_##N##A,
 
+#undef ENTRY
+#define ENTRY(N, M, U, F) \
+  AARCH64_##N,
+
 enum aarch64_builtins
 {
   AARCH64_BUILTIN_MIN,
@@ -829,6 +833,10 @@  enum aarch64_builtins
   AARCH64_RBIT,
   AARCH64_RBITL,
   AARCH64_RBITLL,
+  /* Pragma builtins.  */
+  AARCH64_PRAGMA_BUILTIN_START,
+#include "aarch64-simd-pragma-builtins.def"
+  AARCH64_PRAGMA_BUILTIN_END,
   /* System register builtins.  */
   AARCH64_RSR,
   AARCH64_RSRP,
@@ -947,6 +955,7 @@  const char *aarch64_scalar_builtin_types[] = {
 
 extern GTY(()) aarch64_simd_type_info aarch64_simd_types[];
 
+#undef ENTRY
 #define ENTRY(E, M, Q, G)  \
   {E, "__" #E, #G "__" #E, NULL_TREE, NULL_TREE, E_##M##mode, qualifier_##Q},
 struct aarch64_simd_type_info aarch64_simd_types [] = {
@@ -1547,6 +1556,39 @@  aarch64_init_simd_builtin_functions (bool called_from_pragma)
     }
 }
 
+/* Initialize pragma builtins.  */
+
+typedef struct
+{
+  const char *name;
+  machine_mode mode;
+  int unspec;
+  aarch64_feature_flags required_extensions;
+} pragma_builtins_data;
+
+#undef ENTRY
+#define ENTRY(N, M, U, F) \
+  {#N, E_##M##mode, U, F},
+
+pragma_builtins_data pragma_builtins[] = {
+#include "aarch64-simd-pragma-builtins.def"
+};
+
+static void
+aarch64_init_pragma_builtins ()
+{
+  for (size_t i = 0; i < ARRAY_SIZE (pragma_builtins); ++i)
+    {
+      pragma_builtins_data data = pragma_builtins[i];
+      tree type = aarch64_simd_builtin_type (data.mode, qualifier_none);
+      tree fntype = build_function_type_list (type, type, type, NULL_TREE);
+      unsigned int code = AARCH64_PRAGMA_BUILTIN_START + i + 1;
+      const char *name = data.name;
+      aarch64_builtin_decls[code]
+	= aarch64_general_simulate_builtin (name, fntype, code);
+    }
+}
+
 /* Register the tuple type that contains NUM_VECTORS of the AdvSIMD type
    indexed by TYPE_INDEX.  */
 static void
@@ -1640,6 +1682,7 @@  handle_arm_neon_h (void)
 
   aarch64_init_simd_builtin_functions (true);
   aarch64_init_simd_intrinsics ();
+  aarch64_init_pragma_builtins ();
 }
 
 static void
@@ -2326,6 +2369,16 @@  aarch64_general_check_builtin_call (location_t location, vec<location_t>,
     return aarch64_check_required_extensions (location, decl,
 					      AARCH64_FL_MEMTAG);
 
+  if (code > AARCH64_PRAGMA_BUILTIN_START
+      && code < AARCH64_PRAGMA_BUILTIN_END)
+    {
+      unsigned int pragma_builtins_idx =
+	code - (AARCH64_PRAGMA_BUILTIN_START + 1);
+      aarch64_feature_flags flag =
+	pragma_builtins[pragma_builtins_idx].required_extensions;
+      return aarch64_check_required_extensions (location, decl, flag);
+    }
+
   return true;
 }
 
@@ -3189,6 +3242,29 @@  aarch64_expand_builtin_data_intrinsic (unsigned int fcode, tree exp, rtx target)
   return ops[0].value;
 }
 
+static rtx
+aarch64_expand_pragma_builtin (unsigned int fcode, tree exp, rtx target)
+{
+  machine_mode mode = TYPE_MODE (TREE_TYPE (exp));
+  rtx op0 = force_reg (mode, expand_normal (CALL_EXPR_ARG (exp, 0)));
+  rtx op1 = force_reg (mode, expand_normal (CALL_EXPR_ARG (exp, 1)));
+
+  int unspec;
+  if (fcode >= AARCH64_vamax_f16 && fcode <= AARCH64_vamaxq_f64)
+    unspec = UNSPEC_FAMAX;
+  else if (fcode >= AARCH64_vamin_f16 && fcode <= AARCH64_vaminq_f64)
+    unspec = UNSPEC_FAMIN;
+  else
+    gcc_unreachable ();
+
+  enum insn_code icode = code_for_aarch64 (unspec, mode);
+  rtx pat = GEN_FCN (icode) (target, op0, op1);
+
+  emit_insn (pat);
+
+  return target;
+}
+
 /* Expand an expression EXP as fpsr or fpcr setter (depending on
    UNSPEC) using MODE.  */
 static void
@@ -3368,6 +3444,9 @@  aarch64_general_expand_builtin (unsigned int fcode, tree exp, rtx target,
   if (fcode >= AARCH64_REV16
       && fcode <= AARCH64_RBITLL)
     return aarch64_expand_builtin_data_intrinsic (fcode, exp, target);
+  if (fcode > AARCH64_PRAGMA_BUILTIN_START
+      && fcode < AARCH64_PRAGMA_BUILTIN_END)
+    return aarch64_expand_pragma_builtin (fcode, exp, target);
 
   gcc_unreachable ();
 }
diff --git a/gcc/config/aarch64/aarch64-option-extensions.def b/gcc/config/aarch64/aarch64-option-extensions.def
index 6998627f377..8279f5a76ea 100644
--- a/gcc/config/aarch64/aarch64-option-extensions.def
+++ b/gcc/config/aarch64/aarch64-option-extensions.def
@@ -234,6 +234,8 @@  AARCH64_OPT_EXTENSION("gcs", GCS, (), (), (), "gcs")
 
 AARCH64_OPT_EXTENSION("fp8", FP8, (SIMD), (), (), "fp8")
 
+AARCH64_OPT_EXTENSION("faminmax", FAMINMAX, (SIMD), (), (), "faminmax")
+
 #undef AARCH64_OPT_FMV_EXTENSION
 #undef AARCH64_OPT_EXTENSION
 #undef AARCH64_FMV_FEATURE
diff --git a/gcc/config/aarch64/aarch64-simd-pragma-builtins.def b/gcc/config/aarch64/aarch64-simd-pragma-builtins.def
new file mode 100644
index 00000000000..be7029c4424
--- /dev/null
+++ b/gcc/config/aarch64/aarch64-simd-pragma-builtins.def
@@ -0,0 +1,31 @@ 
+/* AArch64 SIMD pragma builtins
+   Copyright (C) 2024 Free Software Foundation, Inc.
+   Contributed by ARM Ltd.
+
+   This file is part of GCC.
+
+   GCC is free software; you can redistribute it and/or modify it
+   under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3, or (at your option)
+   any later version.
+
+   GCC is distributed in the hope that it will be useful, but
+   WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with GCC; see the file COPYING3.  If not see
+   <http://www.gnu.org/licenses/>.  */
+
+  // faminmax
+  ENTRY (vamax_f16, V4HF, UNSPEC_FAMAX, AARCH64_FL_FAMINMAX)
+  ENTRY (vamaxq_f16, V8HF, UNSPEC_FAMAX, AARCH64_FL_FAMINMAX)
+  ENTRY (vamax_f32, V2SF, UNSPEC_FAMAX, AARCH64_FL_FAMINMAX)
+  ENTRY (vamaxq_f32, V4SF, UNSPEC_FAMAX, AARCH64_FL_FAMINMAX)
+  ENTRY (vamaxq_f64, V2DF, UNSPEC_FAMAX, AARCH64_FL_FAMINMAX)
+  ENTRY (vamin_f16, V4HF, UNSPEC_FAMIN, AARCH64_FL_FAMINMAX)
+  ENTRY (vaminq_f16, V8HF, UNSPEC_FAMIN, AARCH64_FL_FAMINMAX)
+  ENTRY (vamin_f32, V2SF, UNSPEC_FAMIN, AARCH64_FL_FAMINMAX)
+  ENTRY (vaminq_f32, V4SF, UNSPEC_FAMIN, AARCH64_FL_FAMINMAX)
+  ENTRY (vaminq_f64, V2DF, UNSPEC_FAMIN, AARCH64_FL_FAMINMAX)
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 23c03a96371..7542c81ed91 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -9910,3 +9910,14 @@ 
   "shl\\t%d0, %d1, #16"
   [(set_attr "type" "neon_shift_imm")]
 )
+
+;; faminmax
+(define_insn "@aarch64_<faminmax_uns_op><mode>"
+  [(set (match_operand:VHSDF 0 "register_operand" "=w")
+	(unspec:VHSDF [(match_operand:VHSDF 1 "register_operand" "w")
+		       (match_operand:VHSDF 2 "register_operand" "w")]
+		      FAMINMAX_UNS))]
+  "TARGET_FAMINMAX"
+  "<faminmax_uns_op>\t%0.<Vtype>, %1.<Vtype>, %2.<Vtype>"
+  [(set_attr "type" "neon_fp_aminmax<q>")]
+)
diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h
index 2dfb999bea5..de14f57071a 100644
--- a/gcc/config/aarch64/aarch64.h
+++ b/gcc/config/aarch64/aarch64.h
@@ -457,6 +457,10 @@  constexpr auto AARCH64_FL_DEFAULT_ISA_MODE ATTRIBUTE_UNUSED
     enabled through +gcs.  */
 #define TARGET_GCS AARCH64_HAVE_ISA (GCS)
 
+/*  Floating Point Absolute Maximum/Minimum extension instructions are
+    enabled through +faminmax.  */
+#define TARGET_FAMINMAX AARCH64_HAVE_ISA (FAMINMAX)
+
 /* Prefer different predicate registers for the output of a predicated
    operation over re-using an existing input predicate.  */
 #define TARGET_SVE_PRED_CLOBBER (TARGET_SVE \
diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
index 20a318e023b..17ac5e073aa 100644
--- a/gcc/config/aarch64/iterators.md
+++ b/gcc/config/aarch64/iterators.md
@@ -1057,6 +1057,8 @@ 
     UNSPEC_BFCVTN2     ; Used in aarch64-simd.md.
     UNSPEC_BFCVT       ; Used in aarch64-simd.md.
     UNSPEC_FCVTXN	; Used in aarch64-simd.md.
+    UNSPEC_FAMAX       ; Used in aarch64-simd.md.
+    UNSPEC_FAMIN       ; Used in aarch64-simd.md.
 
     ;; All used in aarch64-sve2.md
     UNSPEC_FCVTN
@@ -4463,3 +4465,10 @@ 
    (UNSPECV_SET_FPCR "fpcr")])
 
 (define_int_attr bits_etype [(8 "b") (16 "h") (32 "s") (64 "d")])
+
+;; Iterators and attributes for faminmax
+
+(define_int_iterator FAMINMAX_UNS [UNSPEC_FAMAX UNSPEC_FAMIN])
+
+(define_int_attr faminmax_uns_op
+  [(UNSPEC_FAMAX "famax") (UNSPEC_FAMIN "famin")])
diff --git a/gcc/config/arm/types.md b/gcc/config/arm/types.md
index 9527bdb9e87..d8de9dbc9d1 100644
--- a/gcc/config/arm/types.md
+++ b/gcc/config/arm/types.md
@@ -492,6 +492,8 @@ 
 ; neon_fp_reduc_minmax_s_q
 ; neon_fp_reduc_minmax_d
 ; neon_fp_reduc_minmax_d_q
+; neon_fp_aminmax
+; neon_fp_aminmax_q
 ; neon_fp_cvt_narrow_s_q
 ; neon_fp_cvt_narrow_d_q
 ; neon_fp_cvt_widen_h
@@ -1044,6 +1046,8 @@ 
   neon_fp_reduc_minmax_d,\
   neon_fp_reduc_minmax_d_q,\
 \
+  neon_fp_aminmax,\
+  neon_fp_aminmax_q,\
   neon_fp_cvt_narrow_s_q,\
   neon_fp_cvt_narrow_d_q,\
   neon_fp_cvt_widen_h,\
@@ -1264,6 +1268,8 @@ 
           neon_fp_reduc_add_d_q, neon_fp_reduc_minmax_s,
           neon_fp_reduc_minmax_s_q, neon_fp_reduc_minmax_d,\
           neon_fp_reduc_minmax_d_q,\
+	  neon_fp_aminmax, neon_fp_aminmax_q,\
+          neon_fp_aminmax, neon_fp_aminmax_q,\
           neon_fp_cvt_narrow_s_q, neon_fp_cvt_narrow_d_q,\
           neon_fp_cvt_widen_h, neon_fp_cvt_widen_s, neon_fp_to_int_s,\
           neon_fp_to_int_s_q, neon_int_to_fp_s, neon_int_to_fp_s_q,\
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index 32b772d2a8a..2c509f62d98 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -21865,6 +21865,8 @@  Enable support for Armv8.9-a/9.4-a translation hardening extension.
 Enable the RCpc3 (Release Consistency) extension.
 @item fp8
 Enable the fp8 (8-bit floating point) extension.
+@item faminmax
+Enable the Floating Point Absolute Maximum/Minimum extension.
 
 @end table
 
diff --git a/gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins-no-flag.c b/gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins-no-flag.c
new file mode 100644
index 00000000000..63ed1508c23
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins-no-flag.c
@@ -0,0 +1,10 @@ 
+/* { dg-do assemble} */
+/* { dg-additional-options "-march=armv9-a" } */
+
+#include "arm_neon.h"
+
+void
+test (float32x4_t a, float32x4_t b)
+{
+  vamaxq_f32 (a, b); /* { dg-error {ACLE function 'vamaxq_f32' requires ISA extension 'faminmax'} } */
+}
diff --git a/gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins.c b/gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins.c
new file mode 100644
index 00000000000..7e4f3eba81a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/simd/faminmax-builtins.c
@@ -0,0 +1,115 @@ 
+/* { dg-do assemble} */
+/* { dg-additional-options "-O3 -march=armv9-a+faminmax" } */
+/* { dg-final { check-function-bodies "**" "" } } */
+
+#include "arm_neon.h"
+
+/*
+** test_vamax_f16:
+**	famax	v0.4h, v0.4h, v1.4h
+**	ret
+*/
+float16x4_t
+test_vamax_f16 (float16x4_t a, float16x4_t b)
+{
+  return vamax_f16 (a, b);
+}
+
+/*
+** test_vamaxq_f16:
+**	famax	v0.8h, v0.8h, v1.8h
+**	ret
+*/
+float16x8_t
+test_vamaxq_f16 (float16x8_t a, float16x8_t b)
+{
+  return vamaxq_f16 (a, b);
+}
+
+/*
+** test_vamax_f32:
+**	famax	v0.2s, v0.2s, v1.2s
+**	ret
+*/
+float32x2_t
+test_vamax_f32 (float32x2_t a, float32x2_t b)
+{
+  return vamax_f32 (a, b);
+}
+
+/*
+** test_vamaxq_f32:
+**	famax	v0.4s, v0.4s, v1.4s
+**	ret
+*/
+float32x4_t
+test_vamaxq_f32 (float32x4_t a, float32x4_t b)
+{
+  return vamaxq_f32 (a, b);
+}
+
+/*
+** test_vamaxq_f64:
+**	famax	v0.2d, v0.2d, v1.2d
+**	ret
+*/
+float64x2_t
+test_vamaxq_f64 (float64x2_t a, float64x2_t b)
+{
+  return vamaxq_f64 (a, b);
+}
+
+/*
+** test_vamin_f16:
+**	famin	v0.4h, v0.4h, v1.4h
+**	ret
+*/
+float16x4_t
+test_vamin_f16 (float16x4_t a, float16x4_t b)
+{
+  return vamin_f16 (a, b);
+}
+
+/*
+** test_vaminq_f16:
+**	famin	v0.8h, v0.8h, v1.8h
+**	ret
+*/
+float16x8_t
+test_vaminq_f16 (float16x8_t a, float16x8_t b)
+{
+  return vaminq_f16 (a, b);
+}
+
+/*
+** test_vamin_f32:
+**	famin	v0.2s, v0.2s, v1.2s
+**	ret
+*/
+float32x2_t
+test_vamin_f32 (float32x2_t a, float32x2_t b)
+{
+  return vamin_f32 (a, b);
+}
+
+/*
+** test_vaminq_f32:
+**	famin	v0.4s, v0.4s, v1.4s
+**	ret
+*/
+float32x4_t
+test_vaminq_f32 (float32x4_t a, float32x4_t b)
+{
+  return vaminq_f32 (a, b);
+}
+
+/*
+** test_vaminq_f64:
+**	famin	v0.2d, v0.2d, v1.2d
+**	ret
+*/
+float64x2_t
+test_vaminq_f64 (float64x2_t a, float64x2_t b)
+{
+  return vaminq_f64 (a, b);
+}