[v2,1/1,RISC-V] Add support for _Bfloat16

Message ID 20240402092210.80779-2-zengxiao@eswincomputing.com
State Superseded
Delegated to: Jeff Law
Headers
Series Add support for _Bfloat16 |

Checks

Context Check Description
rivoscibot/toolchain-ci-rivos-lint warning Lint failed
rivoscibot/toolchain-ci-rivos-apply-patch success Patch applied
linaro-tcwg-bot/tcwg_gcc_build--master-aarch64 success Testing passed
rivoscibot/toolchain-ci-rivos-build--newlib-rv64gc-lp64d-multilib success Build passed
rivoscibot/toolchain-ci-rivos-build--linux-rv32gc_zba_zbb_zbc_zbs-ilp32d-non-multilib success Build passed
rivoscibot/toolchain-ci-rivos-build--linux-rv64gc_zba_zbb_zbc_zbs-lp64d-non-multilib success Build passed
rivoscibot/toolchain-ci-rivos-build--newlib-rv64gcv-lp64d-multilib success Build passed
rivoscibot/toolchain-ci-rivos-build--linux-rv64gcv-lp64d-multilib success Build passed
linaro-tcwg-bot/tcwg_gcc_check--master-aarch64 success Testing passed
linaro-tcwg-bot/tcwg_gcc_build--master-arm success Testing passed
linaro-tcwg-bot/tcwg_gcc_check--master-arm success Testing passed
rivoscibot/toolchain-ci-rivos-test success Testing passed

Commit Message

Xiao Zeng April 2, 2024, 9:22 a.m. UTC
  1 At point <https://github.com/riscv/riscv-bfloat16>,
  BF16 has already been completed "post public review".

2 LLVM has also added support for RISCV BF16 in
  <https://reviews.llvm.org/D151313> and
  <https://reviews.llvm.org/D150929>.

3 According to the discussion <https://github.com/riscv-non-isa/riscv-elf-psabi-doc/pull/367>,
  this use __bf16 and use DF16b in riscv_mangle_type like x86.

Below test are passed for this patch
    * The riscv fully regression test.

gcc/ChangeLog:

	* config/riscv/iterators.md: New mode iterator HFBF.
	* config/riscv/riscv-builtins.cc (riscv_init_builtin_types):
	Initialize data type _Bfloat16.
	* config/riscv/riscv-modes.def (FLOAT_MODE): New.
	(ADJUST_FLOAT_FORMAT): New.
	* config/riscv/riscv.cc (riscv_mangle_type): Support for BFmode.
	(riscv_scalar_mode_supported_p): Ditto.
	(riscv_libgcc_floating_mode_supported_p): Ditto.
	(riscv_init_libfuncs): Set the conversion method for BFmode and
	HFmode.
	(riscv_block_arith_comp_libfuncs_for_mode): Set the arithmetic
	and comparison libfuncs for the mode.
	* config/riscv/riscv.md (mode" ): Add BF.
	(movhf): Support for BFmode.
	(mov<mode>): Ditto.
	(*movhf_softfloat): Ditto.
	(*mov<mode>_softfloat): Ditto.

libgcc/ChangeLog:

	* config/riscv/sfp-machine.h (_FP_NANFRAC_B): New.
	(_FP_NANSIGN_B): Ditto.
	* config/riscv/t-softfp32: Add support for BF16 libfuncs.
	* config/riscv/t-softfp64: Ditto.
	* soft-fp/floatsibf.c: For si -> bf16.
	* soft-fp/floatunsibf.c: For unsi -> bf16.

gcc/testsuite/ChangeLog:

	* gcc.target/riscv/bf16_arithmetic.c: New test.
	* gcc.target/riscv/bf16_call.c: New test.
	* gcc.target/riscv/bf16_comparison.c: New test.
	* gcc.target/riscv/bf16_float_libcall_convert.c: New test.
	* gcc.target/riscv/bf16_integer_libcall_convert.c: New test.

Co-authored-by: Jin Ma <jinma@linux.alibaba.com>
---
 gcc/config/riscv/iterators.md                 |  2 +
 gcc/config/riscv/riscv-builtins.cc            | 16 ++++
 gcc/config/riscv/riscv-modes.def              |  3 +
 gcc/config/riscv/riscv.cc                     | 64 ++++++++++-----
 gcc/config/riscv/riscv.md                     | 24 +++---
 .../gcc.target/riscv/bf16_arithmetic.c        | 42 ++++++++++
 gcc/testsuite/gcc.target/riscv/bf16_call.c    | 12 +++
 .../gcc.target/riscv/bf16_comparison.c        | 36 +++++++++
 .../riscv/bf16_float_libcall_convert.c        | 57 +++++++++++++
 .../riscv/bf16_integer_libcall_convert.c      | 81 +++++++++++++++++++
 libgcc/config/riscv/sfp-machine.h             |  3 +
 libgcc/config/riscv/t-softfp32                | 10 ++-
 libgcc/config/riscv/t-softfp64                |  3 +-
 libgcc/soft-fp/floatsibf.c                    | 45 +++++++++++
 libgcc/soft-fp/floatunsibf.c                  | 45 +++++++++++
 15 files changed, 407 insertions(+), 36 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/riscv/bf16_arithmetic.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/bf16_call.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/bf16_comparison.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/bf16_float_libcall_convert.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/bf16_integer_libcall_convert.c
 create mode 100644 libgcc/soft-fp/floatsibf.c
 create mode 100644 libgcc/soft-fp/floatunsibf.c
  

Comments

Jin Ma April 3, 2024, 3:19 a.m. UTC | #1
> gcc/testsuite/ChangeLog:
> 
> 	* gcc.target/riscv/bf16_arithmetic.c: New test.
> 	* gcc.target/riscv/bf16_call.c: New test.
> 	* gcc.target/riscv/bf16_comparison.c: New test.
> 	* gcc.target/riscv/bf16_float_libcall_convert.c: New test.
> 	* gcc.target/riscv/bf16_integer_libcall_convert.c: New test.

  Hi, I have test this patch and it is very good. I think we need to add some
runable tests to ensure that the results are right for various types of
conversions, operations, and libfuncs.

BR,
Jin
  
Palmer Dabbelt April 3, 2024, 3:43 a.m. UTC | #2
On Tue, 02 Apr 2024 20:19:16 PDT (-0700), jinma@linux.alibaba.com wrote:
>> gcc/testsuite/ChangeLog:
>> 
>> 	* gcc.target/riscv/bf16_arithmetic.c: New test.
>> 	* gcc.target/riscv/bf16_call.c: New test.
>> 	* gcc.target/riscv/bf16_comparison.c: New test.
>> 	* gcc.target/riscv/bf16_float_libcall_convert.c: New test.
>> 	* gcc.target/riscv/bf16_integer_libcall_convert.c: New test.
>
>   Hi, I have test this patch and it is very good. I think we need to add some
> runable tests to ensure that the results are right for various types of
> conversions, operations, and libfuncs.

Sorry I forgot to reply earlier, a few of us were talking about this is 
the patchwork meeting this morning.  We think this is too big for GCC-14 
this late in the cycle.  Folks are still looking at bugs, so it might 
take a bit to get reviewed for GCC-15.

I took a look and don't see anything wrong, but I'm not a floating-point 
person so I'd want to try and talk to someone who is before committing 
it.

More testing never hurts, though ;)
>
> BR,
> Jin
  
Xiao Zeng April 3, 2024, 6:16 a.m. UTC | #3
2024-04-03 11:19  Jin Ma <jinma@linux.alibaba.com> wrote:
>
 
>> gcc/testsuite/ChangeLog:
>>
>> * gcc.target/riscv/bf16_arithmetic.c: New test.
>> * gcc.target/riscv/bf16_call.c: New test.
>> * gcc.target/riscv/bf16_comparison.c: New test.
>> * gcc.target/riscv/bf16_float_libcall_convert.c: New test.
>> * gcc.target/riscv/bf16_integer_libcall_convert.c: New test.
>
>  Hi, I have test this patch and it is very good. I think we need to add some
>runable tests to ensure that the results are right for various types of
>conversions, operations, and libfuncs. 
Yes, we must ensure that running tests is also feasible.

A great testcase has already been provided in the gcc test suite: gcc/testsuite/g++.dg/cpp23/ext-floating14.C

So, I didn't add any test cases to run the tests.
>
>BR,
>Jin
 
Thanks
Xiao Zeng
  
Jeff Law May 4, 2024, 3:23 p.m. UTC | #4
On 4/2/24 3:22 AM, Xiao Zeng wrote:
> 1 At point <https://github.com/riscv/riscv-bfloat16>,
>    BF16 has already been completed "post public review".
> 
> 2 LLVM has also added support for RISCV BF16 in
>    <https://reviews.llvm.org/D151313> and
>    <https://reviews.llvm.org/D150929>.
> 
> 3 According to the discussion <https://github.com/riscv-non-isa/riscv-elf-psabi-doc/pull/367>,
>    this use __bf16 and use DF16b in riscv_mangle_type like x86.
> 
> Below test are passed for this patch
>      * The riscv fully regression test.
> 
> gcc/ChangeLog:
> 
> 	* config/riscv/iterators.md: New mode iterator HFBF.
> 	* config/riscv/riscv-builtins.cc (riscv_init_builtin_types):
> 	Initialize data type _Bfloat16.
> 	* config/riscv/riscv-modes.def (FLOAT_MODE): New.
> 	(ADJUST_FLOAT_FORMAT): New.
> 	* config/riscv/riscv.cc (riscv_mangle_type): Support for BFmode.
> 	(riscv_scalar_mode_supported_p): Ditto.
> 	(riscv_libgcc_floating_mode_supported_p): Ditto.
> 	(riscv_init_libfuncs): Set the conversion method for BFmode and
> 	HFmode.
> 	(riscv_block_arith_comp_libfuncs_for_mode): Set the arithmetic
> 	and comparison libfuncs for the mode.
> 	* config/riscv/riscv.md (mode" ): Add BF.
> 	(movhf): Support for BFmode.
> 	(mov<mode>): Ditto.
> 	(*movhf_softfloat): Ditto.
> 	(*mov<mode>_softfloat): Ditto.
> 
> libgcc/ChangeLog:
> 
> 	* config/riscv/sfp-machine.h (_FP_NANFRAC_B): New.
> 	(_FP_NANSIGN_B): Ditto.
> 	* config/riscv/t-softfp32: Add support for BF16 libfuncs.
> 	* config/riscv/t-softfp64: Ditto.
> 	* soft-fp/floatsibf.c: For si -> bf16.
> 	* soft-fp/floatunsibf.c: For unsi -> bf16.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* gcc.target/riscv/bf16_arithmetic.c: New test.
> 	* gcc.target/riscv/bf16_call.c: New test.
> 	* gcc.target/riscv/bf16_comparison.c: New test.
> 	* gcc.target/riscv/bf16_float_libcall_convert.c: New test.
> 	* gcc.target/riscv/bf16_integer_libcall_convert.c: New test.
Just some nits.  In t-softfp32 and t-softfp64 the code you've added 
should be using tabs, not 8 spaces, as noted by the CI "Lint Status":

https://github.com/ewlu/gcc-precommit-ci/issues/1412#issuecomment-2031568644

With that fixed, this is fine for the trunk.  No need to repost, go 
ahead and commit.

Thanks for your patience,
Jeff
  
Xiao Zeng May 5, 2024, 2:08 a.m. UTC | #5
2024-05-04 23:23  Jeff Law <jeffreyalaw@gmail.com> wrote:
>
 
>
>
>On 4/2/24 3:22 AM, Xiao Zeng wrote:
>> 1 At point <https://github.com/riscv/riscv-bfloat16>,
>>    BF16 has already been completed "post public review".
>>
>> 2 LLVM has also added support for RISCV BF16 in
>> <https://reviews.llvm.org/D151313> and
>> <https://reviews.llvm.org/D150929>.
>>
>> 3 According to the discussion <https://github.com/riscv-non-isa/riscv-elf-psabi-doc/pull/367>,
>>    this use __bf16 and use DF16b in riscv_mangle_type like x86.
>>
>> Below test are passed for this patch
>>      * The riscv fully regression test.
>>
>> gcc/ChangeLog:
>>
>> * config/riscv/iterators.md: New mode iterator HFBF.
>> * config/riscv/riscv-builtins.cc (riscv_init_builtin_types):
>> Initialize data type _Bfloat16.
>> * config/riscv/riscv-modes.def (FLOAT_MODE): New.
>> (ADJUST_FLOAT_FORMAT): New.
>> * config/riscv/riscv.cc (riscv_mangle_type): Support for BFmode.
>> (riscv_scalar_mode_supported_p): Ditto.
>> (riscv_libgcc_floating_mode_supported_p): Ditto.
>> (riscv_init_libfuncs): Set the conversion method for BFmode and
>> HFmode.
>> (riscv_block_arith_comp_libfuncs_for_mode): Set the arithmetic
>> and comparison libfuncs for the mode.
>> * config/riscv/riscv.md (mode" ): Add BF.
>> (movhf): Support for BFmode.
>> (mov<mode>): Ditto.
>> (*movhf_softfloat): Ditto.
>> (*mov<mode>_softfloat): Ditto.
>>
>> libgcc/ChangeLog:
>>
>> * config/riscv/sfp-machine.h (_FP_NANFRAC_B): New.
>> (_FP_NANSIGN_B): Ditto.
>> * config/riscv/t-softfp32: Add support for BF16 libfuncs.
>> * config/riscv/t-softfp64: Ditto.
>> * soft-fp/floatsibf.c: For si -> bf16.
>> * soft-fp/floatunsibf.c: For unsi -> bf16.
>>
>> gcc/testsuite/ChangeLog:
>>
>> * gcc.target/riscv/bf16_arithmetic.c: New test.
>> * gcc.target/riscv/bf16_call.c: New test.
>> * gcc.target/riscv/bf16_comparison.c: New test.
>> * gcc.target/riscv/bf16_float_libcall_convert.c: New test.
>> * gcc.target/riscv/bf16_integer_libcall_convert.c: New test.
>Just some nits.  In t-softfp32 and t-softfp64 the code you've added
>should be using tabs, not 8 spaces, as noted by the CI "Lint Status":
>
>https://github.com/ewlu/gcc-precommit-ci/issues/1412#issuecomment-2031568644
In the future, my patch will strictly adhere to the formatting suggestions provided by CI.

>
>With that fixed, this is fine for the trunk.  No need to repost, 
>go ahead and commit.
Currently, I do not have commit permission. Can I have this permission?

>
>Thanks for your patience,
>Jeff
 
Thanks
Xiao Zeng
  
Jeff Law May 6, 2024, 9:06 p.m. UTC | #6
On 5/4/24 8:08 PM, Xiao Zeng wrote:

>>
>> https://github.com/ewlu/gcc-precommit-ci/issues/1412#issuecomment-2031568644
> In the future, my patch will strictly adhere to the formatting suggestions provided by CI.
No worries.  Even those of us who have been working on the project for 
30+ years still goof this stuff up from time to time.   In fact, it 
complained about one of my patches over the weekend ;-)

>>
>> With that fixed, this is fine for the trunk.  No need to repost,
>> go ahead and commit.
> Currently, I do not have commit permission. Can I have this permission?

Use this form:
> https://sourceware.org/cgi-bin/pdw/ps_form.cgi

And my email address as the as your sponsor.  jlaw@ventanamicro.com

I'll go ahead and commit the Bfloat16 patch.  But if you plan on 
contributing regularly, it's definitely easier to have write access.

Jeff
  

Patch

diff --git a/gcc/config/riscv/iterators.md b/gcc/config/riscv/iterators.md
index a7694137685..40bf20f42bb 100644
--- a/gcc/config/riscv/iterators.md
+++ b/gcc/config/riscv/iterators.md
@@ -75,6 +75,8 @@ 
 ;; Iterator for floating-point modes that can be loaded into X registers.
 (define_mode_iterator SOFTF [SF (DF "TARGET_64BIT") (HF "TARGET_ZFHMIN")])
 
+;; Iterator for floating-point modes of BF16
+(define_mode_iterator HFBF [HF BF])
 
 ;; -------------------------------------------------------------------
 ;; Mode attributes
diff --git a/gcc/config/riscv/riscv-builtins.cc b/gcc/config/riscv/riscv-builtins.cc
index d457e306dd1..4c08834288a 100644
--- a/gcc/config/riscv/riscv-builtins.cc
+++ b/gcc/config/riscv/riscv-builtins.cc
@@ -230,6 +230,7 @@  static GTY(()) int riscv_builtin_decl_index[NUM_INSN_CODES];
   riscv_builtin_decls[riscv_builtin_decl_index[(CODE)]]
 
 tree riscv_float16_type_node = NULL_TREE;
+tree riscv_bfloat16_type_node = NULL_TREE;
 
 /* Return the function type associated with function prototype TYPE.  */
 
@@ -273,6 +274,21 @@  riscv_init_builtin_types (void)
   if (!maybe_get_identifier ("_Float16"))
     lang_hooks.types.register_builtin_type (riscv_float16_type_node,
 					    "_Float16");
+
+  /* Provide the _Bfloat16 type and bfloat16_type_node if needed.  */
+  if (!bfloat16_type_node)
+    {
+      riscv_bfloat16_type_node = make_node (REAL_TYPE);
+      TYPE_PRECISION (riscv_bfloat16_type_node) = 16;
+      SET_TYPE_MODE (riscv_bfloat16_type_node, BFmode);
+      layout_type (riscv_bfloat16_type_node);
+    }
+  else
+    riscv_bfloat16_type_node = bfloat16_type_node;
+
+  if (!maybe_get_identifier ("_Bfloat16"))
+    lang_hooks.types.register_builtin_type (riscv_bfloat16_type_node,
+					    "_Bfloat16");
 }
 
 /* Implement TARGET_INIT_BUILTINS.  */
diff --git a/gcc/config/riscv/riscv-modes.def b/gcc/config/riscv/riscv-modes.def
index bdce89b17aa..6de4e440298 100644
--- a/gcc/config/riscv/riscv-modes.def
+++ b/gcc/config/riscv/riscv-modes.def
@@ -21,6 +21,9 @@  along with GCC; see the file COPYING3.  If not see
 
 FLOAT_MODE (HF, 2, ieee_half_format);
 FLOAT_MODE (TF, 16, ieee_quad_format);
+FLOAT_MODE (BF, 2, 0);
+/* Reuse definition from arm.  */
+ADJUST_FLOAT_FORMAT (BF, &arm_bfloat_half_format);
 
 /* Vector modes.  */
 
diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
index fe9976bfffe..0cd3dd8a23d 100644
--- a/gcc/config/riscv/riscv.cc
+++ b/gcc/config/riscv/riscv.cc
@@ -9873,9 +9873,17 @@  riscv_asan_shadow_offset (void)
 static const char *
 riscv_mangle_type (const_tree type)
 {
-  /* Half-precision float, _Float16 is "DF16_".  */
+  /* Half-precision float, _Float16 is "DF16_" and _Bfloat16 is "DF16b".  */
   if (SCALAR_FLOAT_TYPE_P (type) && TYPE_PRECISION (type) == 16)
-    return "DF16_";
+    {
+      if (TYPE_MODE (type) == HFmode)
+	return "DF16_";
+
+      if (TYPE_MODE (type) == BFmode)
+	return "DF16b";
+
+      gcc_unreachable ();
+    }
 
   /* Mangle all vector type for vector extension.  */
   /* The mangle name follows the rule of RVV LLVM
@@ -9896,19 +9904,20 @@  riscv_mangle_type (const_tree type)
 static bool
 riscv_scalar_mode_supported_p (scalar_mode mode)
 {
-  if (mode == HFmode)
+  if (mode == HFmode || mode == BFmode)
     return true;
   else
     return default_scalar_mode_supported_p (mode);
 }
 
 /* Implement TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P - return TRUE
-   if MODE is HFmode, and punt to the generic implementation otherwise.  */
+   if MODE is HFmode or BFmode, and punt to the generic implementation
+   otherwise.  */
 
 static bool
 riscv_libgcc_floating_mode_supported_p (scalar_float_mode mode)
 {
-  if (mode == HFmode)
+  if (mode == HFmode || mode == BFmode)
     return true;
   else
     return default_libgcc_floating_mode_supported_p (mode);
@@ -9959,27 +9968,42 @@  riscv_floatn_mode (int n, bool extended)
   return default_floatn_mode (n, extended);
 }
 
+/* Record that we have no arithmetic or comparison libfuncs for
+   machine_mode MODE.  */
 static void
-riscv_init_libfuncs (void)
+riscv_block_arith_comp_libfuncs_for_mode (machine_mode mode)
 {
-  /* Half-precision float operations.  The compiler handles all operations
-     with NULL libfuncs by converting to SFmode.  */
+  /* Half-precision float or Brain float operations.  The compiler handles all
+     operations with NULL libfuncs by converting to SFmode.  */
 
   /* Arithmetic.  */
-  set_optab_libfunc (add_optab, HFmode, NULL);
-  set_optab_libfunc (sdiv_optab, HFmode, NULL);
-  set_optab_libfunc (smul_optab, HFmode, NULL);
-  set_optab_libfunc (neg_optab, HFmode, NULL);
-  set_optab_libfunc (sub_optab, HFmode, NULL);
+  set_optab_libfunc (add_optab, mode, NULL);
+  set_optab_libfunc (sdiv_optab, mode, NULL);
+  set_optab_libfunc (smul_optab, mode, NULL);
+  set_optab_libfunc (neg_optab, mode, NULL);
+  set_optab_libfunc (sub_optab, mode, NULL);
 
   /* Comparisons.  */
-  set_optab_libfunc (eq_optab, HFmode, NULL);
-  set_optab_libfunc (ne_optab, HFmode, NULL);
-  set_optab_libfunc (lt_optab, HFmode, NULL);
-  set_optab_libfunc (le_optab, HFmode, NULL);
-  set_optab_libfunc (ge_optab, HFmode, NULL);
-  set_optab_libfunc (gt_optab, HFmode, NULL);
-  set_optab_libfunc (unord_optab, HFmode, NULL);
+  set_optab_libfunc (eq_optab, mode, NULL);
+  set_optab_libfunc (ne_optab, mode, NULL);
+  set_optab_libfunc (lt_optab, mode, NULL);
+  set_optab_libfunc (le_optab, mode, NULL);
+  set_optab_libfunc (ge_optab, mode, NULL);
+  set_optab_libfunc (gt_optab, mode, NULL);
+  set_optab_libfunc (unord_optab, mode, NULL);
+}
+
+static void
+riscv_init_libfuncs (void)
+{
+  riscv_block_arith_comp_libfuncs_for_mode (HFmode);
+  riscv_block_arith_comp_libfuncs_for_mode (BFmode);
+
+  /* Convert between BFmode and HFmode using only trunc libfunc if needed.  */
+  set_conv_libfunc (sext_optab, BFmode, HFmode, "__trunchfbf2");
+  set_conv_libfunc (sext_optab, HFmode, BFmode, "__truncbfhf2");
+  set_conv_libfunc (trunc_optab, BFmode, HFmode, "__trunchfbf2");
+  set_conv_libfunc (trunc_optab, HFmode, BFmode, "__truncbfhf2");
 }
 
 #if CHECKING_P
diff --git a/gcc/config/riscv/riscv.md b/gcc/config/riscv/riscv.md
index 0346cc3859d..3d266ad5c41 100644
--- a/gcc/config/riscv/riscv.md
+++ b/gcc/config/riscv/riscv.md
@@ -197,7 +197,7 @@ 
   (const_string "unknown"))
 
 ;; Main data type used by the insn
-(define_attr "mode" "unknown,none,QI,HI,SI,DI,TI,HF,SF,DF,TF,
+(define_attr "mode" "unknown,none,QI,HI,SI,DI,TI,HF,BF,SF,DF,TF,
   RVVMF64BI,RVVMF32BI,RVVMF16BI,RVVMF8BI,RVVMF4BI,RVVMF2BI,RVVM1BI,
   RVVM8QI,RVVM4QI,RVVM2QI,RVVM1QI,RVVMF2QI,RVVMF4QI,RVVMF8QI,
   RVVM8HI,RVVM4HI,RVVM2HI,RVVM1HI,RVVMF2HI,RVVMF4HI,
@@ -1921,12 +1921,12 @@ 
    (set_attr "mode" "DF")])
 
 ;; 16-bit floating point moves
-(define_expand "movhf"
-  [(set (match_operand:HF 0 "")
-	(match_operand:HF 1 ""))]
+(define_expand "mov<mode>"
+  [(set (match_operand:HFBF 0 "")
+	(match_operand:HFBF 1 ""))]
   ""
 {
-  if (riscv_legitimize_move (HFmode, operands[0], operands[1]))
+  if (riscv_legitimize_move (<MODE>mode, operands[0], operands[1]))
     DONE;
 })
 
@@ -1941,16 +1941,16 @@ 
    (set_attr "type" "fmove,fmove,mtc,fpload,fpstore,store,mtc,mfc,move,load,store")
    (set_attr "mode" "HF")])
 
-(define_insn "*movhf_softfloat"
-  [(set (match_operand:HF 0 "nonimmediate_operand" "=f, r,r,m,*f,*r")
-	(match_operand:HF 1 "move_operand"         " f,Gr,m,r,*r,*f"))]
-  "!TARGET_ZFHMIN
-   && (register_operand (operands[0], HFmode)
-       || reg_or_0_operand (operands[1], HFmode))"
+(define_insn "*mov<mode>_softfloat"
+  [(set (match_operand:HFBF 0 "nonimmediate_operand" "=f, r,r,m,*f,*r")
+	(match_operand:HFBF 1 "move_operand"         " f,Gr,m,r,*r,*f"))]
+  "((!TARGET_ZFHMIN && <MODE>mode == HFmode) || (<MODE>mode == BFmode))
+   && (register_operand (operands[0], <MODE>mode)
+       || reg_or_0_operand (operands[1], <MODE>mode))"
   { return riscv_output_move (operands[0], operands[1]); }
   [(set_attr "move_type" "fmove,move,load,store,mtc,mfc")
    (set_attr "type" "fmove,move,load,store,mtc,mfc")
-   (set_attr "mode" "HF")])
+   (set_attr "mode" "<MODE>")])
 
 (define_insn "*movhf_softfloat_boxing"
   [(set (match_operand:HF 0 "register_operand"            "=f")
diff --git a/gcc/testsuite/gcc.target/riscv/bf16_arithmetic.c b/gcc/testsuite/gcc.target/riscv/bf16_arithmetic.c
new file mode 100644
index 00000000000..9e485051260
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/bf16_arithmetic.c
@@ -0,0 +1,42 @@ 
+/* { dg-do compile } */
+/* { dg-options "-march=rv32i -mabi=ilp32 -O" { target { rv32 } } } */
+/* { dg-options "-march=rv64i -mabi=lp64 -O" { target { rv64 } } } */
+
+/* 1) bf -> sf          (call      __extendbfsf2)  */
+/* 2) sf1 [+|-|*|/] sf2 (call      __[add|sub|mul|div]sf3)  */
+/* 3) sf -> bf          (call      __truncsfbf2)  */
+extern _Bfloat16 bf;
+extern _Bfloat16 bf1;
+extern _Bfloat16 bf2;
+
+void bf_add_bf () { bf = bf1 + bf2; }
+
+void bf_sub_bf () { bf = bf1 - bf2; }
+
+void bf_mul_bf () { bf = bf1 * bf2; }
+
+void bf_div_bf () { bf = bf1 / bf2; }
+
+void bf_add_const () { bf = bf1 + 3.14f; }
+
+void const_sub_bf () { bf = 3.14f - bf2; }
+
+void bf_mul_const () { bf = bf1 * 3.14f; }
+
+void const_div_bf () { bf = 3.14f / bf2; }
+
+void bf_inc () { ++bf; }
+
+void bf_dec () { --bf; }
+
+/* { dg-final { scan-assembler-times "call\t__extendbfsf2" 16 } } */
+/* { dg-final { scan-assembler-times "call\t__truncsfbf2" 10 } } */
+/* { dg-final { scan-assembler-times "call\t__addsf3" 3 } } */
+/* { dg-final { scan-assembler-times "call\t__subsf3" 3 } } */
+/* { dg-final { scan-assembler-times "call\t__mulsf3" 2 } } */
+/* { dg-final { scan-assembler-times "call\t__divsf3" 2 } } */
+
+/* { dg-final { scan-assembler-not "call\t__addbf3" } } */
+/* { dg-final { scan-assembler-not "call\t__subbf3" } } */
+/* { dg-final { scan-assembler-not "call\t__mulbf3" } } */
+/* { dg-final { scan-assembler-not "call\t__divbf3" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/bf16_call.c b/gcc/testsuite/gcc.target/riscv/bf16_call.c
new file mode 100644
index 00000000000..29f67719999
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/bf16_call.c
@@ -0,0 +1,12 @@ 
+/* { dg-do compile } */
+/* { dg-options "-march=rv32i -mabi=ilp32 -O" { target { rv32 } } } */
+/* { dg-options "-march=rv64i -mabi=lp64 -O" { target { rv64 } } } */
+
+/* 1) bf -> sf (call      __extendbfsf2)  */
+/* 2) sf -> bf (call      __truncsfbf2)  */
+__attribute__ ((noinline)) _Bfloat16 add (_Bfloat16 a, _Bfloat16 b) { return a + b; }
+
+_Bfloat16 test(_Bfloat16 a, _Bfloat16 b) { return add (a, b); }
+
+/* { dg-final { scan-assembler-times "call\t__extendbfsf2" 2 } } */
+/* { dg-final { scan-assembler-times "call\t__truncsfbf2" 1 } } */
diff --git a/gcc/testsuite/gcc.target/riscv/bf16_comparison.c b/gcc/testsuite/gcc.target/riscv/bf16_comparison.c
new file mode 100644
index 00000000000..69db803f403
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/bf16_comparison.c
@@ -0,0 +1,36 @@ 
+/* { dg-do compile } */
+/* { dg-options "-march=rv32i -mabi=ilp32 -O" { target { rv32 } } } */
+/* { dg-options "-march=rv64i -mabi=lp64 -O" { target { rv64 } } } */
+
+/* 1) bf -> sf               (call	__extendbfsf2)  */
+/* 2) sf1 [<|<=|>|>=|==] sf2 (call	__[lt|le|gt|ge|eq]sf2)  */
+extern _Bfloat16 bf;
+extern _Bfloat16 bf1;
+extern _Bfloat16 bf2;
+
+void bf_lt_bf () { bf = (bf1 < bf2) ? bf1 : bf2; }
+
+void bf_le_bf () { bf = (bf1 <= bf2) ? bf1 : bf2; }
+
+void bf_gt_bf () { bf = (bf1 > bf2) ? bf1 : bf2; }
+
+void bf_ge_bf () { bf = (bf1 >= bf2) ? bf1 : bf2; }
+
+void bf_eq_bf () { bf = (bf1 == bf2) ? bf1 : bf2; }
+
+void bf_lt_const () { bf = (bf1 < 3.14f) ? bf1 : bf2; }
+
+void bf_le_const () { bf = (bf1 <= 3.14f) ? bf1 : bf2; }
+
+void const_gt_bf () { bf = (3.14f > bf2) ? bf1 : bf2; }
+
+void const_ge_bf () { bf = (3.14f >= bf2) ? bf1 : bf2; }
+
+void bf_eq_const () { bf = (bf1 == 3.14f) ? bf1 : bf2; }
+
+/* { dg-final { scan-assembler-times "call\t__extendbfsf2" 15 } } */
+/* { dg-final { scan-assembler-not "call\t__ltbf2" } } */
+/* { dg-final { scan-assembler-not "call\t__lebf2" } } */
+/* { dg-final { scan-assembler-not "call\t__gtbf2" } } */
+/* { dg-final { scan-assembler-not "call\t__gebf2" } } */
+/* { dg-final { scan-assembler-not "call\t__eqbf2" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/bf16_float_libcall_convert.c b/gcc/testsuite/gcc.target/riscv/bf16_float_libcall_convert.c
new file mode 100644
index 00000000000..ba6c6460bc2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/bf16_float_libcall_convert.c
@@ -0,0 +1,57 @@ 
+/* { dg-do compile } */
+/* { dg-options "-march=rv32i -mabi=ilp32 -O" { target { rv32 } } } */
+/* { dg-options "-march=rv64i -mabi=lp64 -O" { target { rv64 } } } */
+
+
+/* 1) float       -> BF16
+ *    hf/sf/df/tf -> bf                   (call   __trunc[h|s|d|t]fbf2)
+*/
+
+/* 2) BF16        -> float
+ *    bf          -> hf == sf -> hf       (call   __truncsfhf2)
+ *    bf          -> sf                   (call   __extendbfsf2)
+ *    bf          -> df == bf -> sf -> df (call   __extendbfsf2 && __extendsfdf2)
+ *    bf          -> tf == bf -> sf -> tf (call   __extendbfsf2 && __extendsftf2)
+*/
+
+extern   _Bfloat16 bf;
+extern   _Float16  hf;
+extern      float  sf;
+extern      double df;
+extern long double tf;
+
+void hf_to_bf () { bf = hf; }
+void bf_to_hf () { hf = bf; }
+
+void sf_to_bf () { bf = sf; }
+void bf_to_sf () { sf = bf; }
+
+void df_to_bf () { bf = df; } 
+void bf_to_df () { df = bf; }
+
+void tf_to_bf () { bf = tf; }
+void bf_to_tf () { tf = bf; }
+
+/* { dg-final { scan-assembler-times "call\t__extendbfsf2" 3 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__extendbfsf2" 3 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__extendsfdf2" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__extendsfdf2" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__extendsftf2" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__extendsftf2" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__truncsfhf2" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__truncsfhf2" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__trunchfbf2" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__trunchfbf2" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__truncsfbf2" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__truncsfbf2" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__truncdfbf2" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__truncdfbf2" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__trunctfbf2" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__trunctfbf2" 1 { target { rv64 } } } } */
diff --git a/gcc/testsuite/gcc.target/riscv/bf16_integer_libcall_convert.c b/gcc/testsuite/gcc.target/riscv/bf16_integer_libcall_convert.c
new file mode 100644
index 00000000000..ad714253a4a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/bf16_integer_libcall_convert.c
@@ -0,0 +1,81 @@ 
+/* { dg-do compile } */
+/* { dg-options "-march=rv32i -mabi=ilp32 -O" { target { rv32 } } } */
+/* { dg-options "-march=rv64i -mabi=lp64 -O" { target { rv64 } } } */
+
+/* 1) Integer     -> BF16 
+ *    qi/hi/si    -> bf (call	__floatsibf) 
+ *    uqi/uhi/usi -> bf (call	__floatunsibf)  
+ *    di/ti       -> bf (call	__float[d|t]ibf)
+ *    udi/uti     -> bf (call	__floatun[d|t]ibf)
+*/
+
+/* 2) BF16        -> Integer
+ *    bf          -> qi/hi/si    (call   __fixsfsi)
+ *    bf          -> uqi/uhi/usi (call   __fixunssfsi)
+ *    bf          -> di/ti       (call   __fixsf[d|t]i)
+ *    bf          -> udi/uti     (call   __fixunssf[d|t]i)
+*/
+
+extern _Bfloat16 bf;
+
+extern   signed char       qi;
+extern unsigned char      uqi;
+extern   signed short      hi;
+extern unsigned short     uhi;
+extern   signed int        si;
+extern unsigned int       usi;
+extern   signed long long  di;
+extern unsigned long long udi;
+
+void qi_to_bf ()  { bf = qi; }
+void uqi_to_bf () { bf = uqi; }
+void bf_to_qi ()  { qi = bf; }
+void bf_to_uqi () { uqi = bf; }
+
+void hi_to_bf ()  { bf = hi; }
+void uhi_to_bf () { bf = uhi; }
+void bf_to_hi ()  { hi = bf; }
+void bf_to_uhi () { uhi = bf; }
+
+void si_to_bf ()  { bf = si; }
+void usi_to_bf () { bf = usi; }
+void bf_to_si ()  { si = bf; }
+void bf_to_usi () { usi = bf; }
+
+void di_to_bf ()  { bf = di; }
+void udi_to_bf () { bf = udi; }
+void bf_to_di ()  { di = bf; }
+void bf_to_udi () { udi = bf; }
+
+#if __riscv_xlen == 64
+extern   signed __int128   ti; 
+extern unsigned __int128  uti; 
+void ti_to_bf ()  { bf = ti; }  /* { dg-final { scan-assembler-times "call\t__floattibf" 1 { target { rv64 } } } } */
+void uti_to_bf () { bf = uti; } /* { dg-final { scan-assembler-times "call\t__floatuntibf" 1 { target { rv64 } } } } */
+void bf_to_ti ()  { ti = bf; }  /* { dg-final { scan-assembler-times "call\t__fixsfti" 1 { target { rv64 } } } } */
+void bf_to_uti () { uti = bf; } /* { dg-final { scan-assembler-times "call\t__fixunssfti" 1 { target { rv64 } } } } */
+#endif
+
+/* { dg-final { scan-assembler-times "call\t__fixsfsi" 3 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__fixsfsi" 3 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__fixsfdi" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__fixsfdi" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__fixunssfsi" 3 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__fixunssfsi" 3 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__fixunssfdi" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__fixunssfdi" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__floatsibf" 3 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__floatsibf" 3 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__floatdibf" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__floatdibf" 1 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__floatunsibf" 3 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__floatunsibf" 3 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "call\t__floatundibf" 1 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "call\t__floatundibf" 1 { target { rv64 } } } } */
diff --git a/libgcc/config/riscv/sfp-machine.h b/libgcc/config/riscv/sfp-machine.h
index 0ad97081e88..9d3552ce48f 100644
--- a/libgcc/config/riscv/sfp-machine.h
+++ b/libgcc/config/riscv/sfp-machine.h
@@ -41,6 +41,7 @@  see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
 #define _FP_DIV_MEAT_D(R,X,Y)	_FP_DIV_MEAT_2_udiv(D,R,X,Y)
 #define _FP_DIV_MEAT_Q(R,X,Y)	_FP_DIV_MEAT_4_udiv(Q,R,X,Y)
 
+#define _FP_NANFRAC_B		_FP_QNANBIT_B
 #define _FP_NANFRAC_H		_FP_QNANBIT_H
 #define _FP_NANFRAC_S		_FP_QNANBIT_S
 #define _FP_NANFRAC_D		_FP_QNANBIT_D, 0
@@ -64,6 +65,7 @@  see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
 #define _FP_DIV_MEAT_D(R,X,Y)	_FP_DIV_MEAT_1_udiv_norm(D,R,X,Y)
 #define _FP_DIV_MEAT_Q(R,X,Y)	_FP_DIV_MEAT_2_udiv(Q,R,X,Y)
 
+#define _FP_NANFRAC_B		_FP_QNANBIT_B
 #define _FP_NANFRAC_H		_FP_QNANBIT_H
 #define _FP_NANFRAC_S		_FP_QNANBIT_S
 #define _FP_NANFRAC_D		_FP_QNANBIT_D
@@ -82,6 +84,7 @@  typedef unsigned int UTItype __attribute__ ((mode (TI)));
 typedef int __gcc_CMPtype __attribute__ ((mode (__libgcc_cmp_return__)));
 #define CMPtype __gcc_CMPtype
 
+#define _FP_NANSIGN_B		0
 #define _FP_NANSIGN_H		0
 #define _FP_NANSIGN_S		0
 #define _FP_NANSIGN_D		0
diff --git a/libgcc/config/riscv/t-softfp32 b/libgcc/config/riscv/t-softfp32
index 1a3b1caa6b0..5e2442fb586 100644
--- a/libgcc/config/riscv/t-softfp32
+++ b/libgcc/config/riscv/t-softfp32
@@ -42,7 +42,11 @@  softfp_extras += divsf3 divdf3 divtf3
 
 endif
 
-softfp_extensions += hfsf hfdf hftf
-softfp_truncations += tfhf dfhf sfhf
+softfp_extensions += hfsf hfdf hftf \
+                     bfsf
+softfp_truncations += tfhf dfhf sfhf \
+                      tfbf dfbf sfbf \
+                      hfbf bfhf
 softfp_extras += fixhfsi fixhfdi fixunshfsi fixunshfdi \
-                 floatsihf floatdihf floatunsihf floatundihf
+                 floatsihf floatdihf floatunsihf floatundihf \
+                 floatsibf floatdibf floatunsibf floatundibf
diff --git a/libgcc/config/riscv/t-softfp64 b/libgcc/config/riscv/t-softfp64
index c87d242d5c3..6078bed439a 100644
--- a/libgcc/config/riscv/t-softfp64
+++ b/libgcc/config/riscv/t-softfp64
@@ -1,4 +1,5 @@ 
 include $(srcdir)/config/riscv/t-softfp32
 
 softfp_int_modes += ti
-softfp_extras += fixhfti fixunshfti floattihf floatuntihf
+softfp_extras += fixhfti fixunshfti floattihf floatuntihf \
+                 floattibf floatuntibf
diff --git a/libgcc/soft-fp/floatsibf.c b/libgcc/soft-fp/floatsibf.c
new file mode 100644
index 00000000000..0cb05e58fc1
--- /dev/null
+++ b/libgcc/soft-fp/floatsibf.c
@@ -0,0 +1,45 @@ 
+/* Software floating-point emulation.
+   Convert a 32bit signed integer to bfloat16
+   Copyright (C) 2024 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include "soft-fp.h"
+#include "brain.h"
+
+BFtype
+__floatsibf (SItype i)
+{
+  FP_DECL_EX;
+  FP_DECL_B (A);
+  BFtype a;
+
+  FP_INIT_ROUNDMODE;
+  FP_FROM_INT_B (A, i, SI_BITS, USItype);
+  FP_PACK_RAW_B (a, A);
+  FP_HANDLE_EXCEPTIONS;
+
+  return a;
+}
diff --git a/libgcc/soft-fp/floatunsibf.c b/libgcc/soft-fp/floatunsibf.c
new file mode 100644
index 00000000000..cd8c4bc8839
--- /dev/null
+++ b/libgcc/soft-fp/floatunsibf.c
@@ -0,0 +1,45 @@ 
+/* Software floating-point emulation.
+   Convert a 32bit unsigned integer to bfloat16
+   Copyright (C) 2007-2024 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   In addition to the permissions in the GNU Lesser General Public
+   License, the Free Software Foundation gives you unlimited
+   permission to link the compiled version of this file into
+   combinations with other programs, and to distribute those
+   combinations without any restriction coming from the use of this
+   file.  (The Lesser General Public License restrictions do apply in
+   other respects; for example, they cover modification of the file,
+   and distribution when not linked into a combine executable.)
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include "soft-fp.h"
+#include "brain.h"
+
+BFtype
+__floatunsibf (USItype i)
+{
+  FP_DECL_EX;
+  FP_DECL_B (A);
+  BFtype a;
+
+  FP_INIT_ROUNDMODE;
+  FP_FROM_INT_B (A, i, SI_BITS, USItype);
+  FP_PACK_RAW_B (a, A);
+  FP_HANDLE_EXCEPTIONS;
+
+  return a;
+}