[v1] RISC-V: Support RVV VFNCVT.X.F.W rounding mode intrinsic API
Checks
Commit Message
From: Pan Li <pan2.li@intel.com>
This patch would like to support the rounding mode API for the
VFNCVT.X.F.W as the below samples.
* __riscv_vfncvt_x_f_w_i16mf2_rm
* __riscv_vfncvt_x_f_w_i16mf2_rm_m
Signed-off-by: Pan Li <pan2.li@intel.com>
gcc/ChangeLog:
* config/riscv/riscv-vector-builtins-bases.cc
(class vfncvt_x): Add frm_op_type template arg.
(BASE): New declaration.
* config/riscv/riscv-vector-builtins-bases.h: Ditto.
* config/riscv/riscv-vector-builtins-functions.def
(vfncvt_x_frm): New intrinsic function def.
* config/riscv/riscv-vector-builtins-shapes.cc
(struct narrow_alu_frm_def): New shape function for frm.
(SHAPE): New declaration.
* config/riscv/riscv-vector-builtins-shapes.h: Ditto.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/base/float-point-ncvt-x.c: New test.
---
.../riscv/riscv-vector-builtins-bases.cc | 9 ++++-
.../riscv/riscv-vector-builtins-bases.h | 1 +
.../riscv/riscv-vector-builtins-functions.def | 2 +
.../riscv/riscv-vector-builtins-shapes.cc | 39 +++++++++++++++++++
.../riscv/riscv-vector-builtins-shapes.h | 1 +
.../riscv/rvv/base/float-point-ncvt-x.c | 29 ++++++++++++++
6 files changed, 80 insertions(+), 1 deletion(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/base/float-point-ncvt-x.c
Comments
Committed, thanks Kito.
Pan
-----Original Message-----
From: Kito Cheng <kito.cheng@gmail.com>
Sent: Thursday, August 17, 2023 3:30 PM
To: Li, Pan2 <pan2.li@intel.com>
Cc: juzhe.zhong@rivai.ai
Subject: Re: [PATCH v1] RISC-V: Support RVV VFNCVT.X.F.W rounding mode intrinsic API
Yeah, I missed that, LGTM :P
On Thu, Aug 17, 2023 at 2:28 PM Li, Pan2 <pan2.li@intel.com> wrote:
>
> Hi Kito,
>
> In case you missed this one, which is the precondition of the rest rounding mode API patches for committing.
> Thank in advance, and we are close to complete all the rounding mode API, 😉.
>
> Pan
>
> -----Original Message-----
> From: Li, Pan2 <pan2.li@intel.com>
> Sent: Wednesday, August 16, 2023 8:54 PM
> To: gcc-patches@gcc.gnu.org
> Cc: juzhe.zhong@rivai.ai; Li, Pan2 <pan2.li@intel.com>; Wang, Yanzhang <yanzhang.wang@intel.com>; kito.cheng@gmail.com
> Subject: [PATCH v1] RISC-V: Support RVV VFNCVT.X.F.W rounding mode intrinsic API
>
> From: Pan Li <pan2.li@intel.com>
>
> This patch would like to support the rounding mode API for the
> VFNCVT.X.F.W as the below samples.
>
> * __riscv_vfncvt_x_f_w_i16mf2_rm
> * __riscv_vfncvt_x_f_w_i16mf2_rm_m
>
> Signed-off-by: Pan Li <pan2.li@intel.com>
>
> gcc/ChangeLog:
>
> * config/riscv/riscv-vector-builtins-bases.cc
> (class vfncvt_x): Add frm_op_type template arg.
> (BASE): New declaration.
> * config/riscv/riscv-vector-builtins-bases.h: Ditto.
> * config/riscv/riscv-vector-builtins-functions.def
> (vfncvt_x_frm): New intrinsic function def.
> * config/riscv/riscv-vector-builtins-shapes.cc
> (struct narrow_alu_frm_def): New shape function for frm.
> (SHAPE): New declaration.
> * config/riscv/riscv-vector-builtins-shapes.h: Ditto.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/riscv/rvv/base/float-point-ncvt-x.c: New test.
> ---
> .../riscv/riscv-vector-builtins-bases.cc | 9 ++++-
> .../riscv/riscv-vector-builtins-bases.h | 1 +
> .../riscv/riscv-vector-builtins-functions.def | 2 +
> .../riscv/riscv-vector-builtins-shapes.cc | 39 +++++++++++++++++++
> .../riscv/riscv-vector-builtins-shapes.h | 1 +
> .../riscv/rvv/base/float-point-ncvt-x.c | 29 ++++++++++++++
> 6 files changed, 80 insertions(+), 1 deletion(-)
> create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/base/float-point-ncvt-x.c
>
> diff --git a/gcc/config/riscv/riscv-vector-builtins-bases.cc b/gcc/config/riscv/riscv-vector-builtins-bases.cc
> index 050ecbe780c..2f40eeaeda5 100644
> --- a/gcc/config/riscv/riscv-vector-builtins-bases.cc
> +++ b/gcc/config/riscv/riscv-vector-builtins-bases.cc
> @@ -1759,10 +1759,15 @@ public:
> };
>
> /* Implements vfncvt.x. */
> -template<int UNSPEC>
> +template<int UNSPEC, enum frm_op_type FRM_OP = NO_FRM>
> class vfncvt_x : public function_base
> {
> public:
> + bool has_rounding_mode_operand_p () const override
> + {
> + return FRM_OP == HAS_FRM;
> + }
> +
> rtx expand (function_expander &e) const override
> {
> return e.use_exact_insn (
> @@ -2502,6 +2507,7 @@ static CONSTEXPR const vfwcvt_rtz_x<FIX> vfwcvt_rtz_x_obj;
> static CONSTEXPR const vfwcvt_rtz_x<UNSIGNED_FIX> vfwcvt_rtz_xu_obj;
> static CONSTEXPR const vfwcvt_f vfwcvt_f_obj;
> static CONSTEXPR const vfncvt_x<UNSPEC_VFCVT> vfncvt_x_obj;
> +static CONSTEXPR const vfncvt_x<UNSPEC_VFCVT, HAS_FRM> vfncvt_x_frm_obj;
> static CONSTEXPR const vfncvt_x<UNSPEC_UNSIGNED_VFCVT> vfncvt_xu_obj;
> static CONSTEXPR const vfncvt_rtz_x<FIX> vfncvt_rtz_x_obj;
> static CONSTEXPR const vfncvt_rtz_x<UNSIGNED_FIX> vfncvt_rtz_xu_obj;
> @@ -2756,6 +2762,7 @@ BASE (vfwcvt_rtz_x)
> BASE (vfwcvt_rtz_xu)
> BASE (vfwcvt_f)
> BASE (vfncvt_x)
> +BASE (vfncvt_x_frm)
> BASE (vfncvt_xu)
> BASE (vfncvt_rtz_x)
> BASE (vfncvt_rtz_xu)
> diff --git a/gcc/config/riscv/riscv-vector-builtins-bases.h b/gcc/config/riscv/riscv-vector-builtins-bases.h
> index 6565740c597..edff0de2715 100644
> --- a/gcc/config/riscv/riscv-vector-builtins-bases.h
> +++ b/gcc/config/riscv/riscv-vector-builtins-bases.h
> @@ -220,6 +220,7 @@ extern const function_base *const vfwcvt_rtz_x;
> extern const function_base *const vfwcvt_rtz_xu;
> extern const function_base *const vfwcvt_f;
> extern const function_base *const vfncvt_x;
> +extern const function_base *const vfncvt_x_frm;
> extern const function_base *const vfncvt_xu;
> extern const function_base *const vfncvt_rtz_x;
> extern const function_base *const vfncvt_rtz_xu;
> diff --git a/gcc/config/riscv/riscv-vector-builtins-functions.def b/gcc/config/riscv/riscv-vector-builtins-functions.def
> index 22c039c8cbb..5e37bae318a 100644
> --- a/gcc/config/riscv/riscv-vector-builtins-functions.def
> +++ b/gcc/config/riscv/riscv-vector-builtins-functions.def
> @@ -472,6 +472,8 @@ DEF_RVV_FUNCTION (vfncvt_f, narrow_alu, full_preds, u_to_nf_xu_w_ops)
> DEF_RVV_FUNCTION (vfncvt_f, narrow_alu, full_preds, f_to_nf_f_w_ops)
> DEF_RVV_FUNCTION (vfncvt_rod_f, narrow_alu, full_preds, f_to_nf_f_w_ops)
>
> +DEF_RVV_FUNCTION (vfncvt_x_frm, narrow_alu_frm, full_preds, f_to_ni_f_w_ops)
> +
> /* 14. Vector Reduction Operations. */
>
> // 14.1. Vector Single-Width Integer Reduction Instructions
> diff --git a/gcc/config/riscv/riscv-vector-builtins-shapes.cc b/gcc/config/riscv/riscv-vector-builtins-shapes.cc
> index 1d14fa21e81..80329113af3 100644
> --- a/gcc/config/riscv/riscv-vector-builtins-shapes.cc
> +++ b/gcc/config/riscv/riscv-vector-builtins-shapes.cc
> @@ -333,6 +333,44 @@ struct widen_alu_frm_def : public build_frm_base
> }
> };
>
> +/* narrow_alu_frm_def class. */
> +struct narrow_alu_frm_def : public build_frm_base
> +{
> + char *get_name (function_builder &b, const function_instance &instance,
> + bool overloaded_p) const override
> + {
> + char base_name[BASE_NAME_MAX_LEN] = {};
> +
> + normalize_base_name (base_name, instance.base_name, sizeof (base_name));
> +
> + b.append_base_name (base_name);
> +
> + if (!overloaded_p)
> + {
> + /* vop --> vop_<op>. */
> + b.append_name (operand_suffixes[instance.op_info->op]);
> + /* vop_<op> --> vop_<op>_<type>. */
> + vector_type_index ret_type_idx
> + = instance.op_info->ret.get_function_type_index (instance.type.index);
> + b.append_name (type_suffixes[ret_type_idx].vector);
> + }
> +
> + /* According to rvv-intrinsic-doc, it does not add "_rm" suffix
> + for vop_rm C++ overloaded API. */
> + if (!overloaded_p)
> + b.append_name ("_rm");
> +
> + /* According to rvv-intrinsic-doc, it does not add "_m" suffix
> + for vop_m C++ overloaded API. */
> + if (overloaded_p && instance.pred == PRED_TYPE_m)
> + return b.finish_name ();
> +
> + b.append_name (predication_suffixes[instance.pred]);
> +
> + return b.finish_name ();
> + }
> +};
> +
> /* widen_alu_def class. Handle vwadd/vwsub. Unlike
> vadd.vx/vadd.vv/vwmul.vv/vwmul.vx, vwadd.vv/vwadd.vx/vwadd.wv/vwadd.wx has
> 'OP' suffix in overloaded API. */
> @@ -856,6 +894,7 @@ SHAPE(widen_alu_frm, widen_alu_frm)
> SHAPE(no_mask_policy, no_mask_policy)
> SHAPE(return_mask, return_mask)
> SHAPE(narrow_alu, narrow_alu)
> +SHAPE(narrow_alu_frm, narrow_alu_frm)
> SHAPE(move, move)
> SHAPE(mask_alu, mask_alu)
> SHAPE(reduc_alu, reduc_alu)
> diff --git a/gcc/config/riscv/riscv-vector-builtins-shapes.h b/gcc/config/riscv/riscv-vector-builtins-shapes.h
> index 841b930b547..b53ab451902 100644
> --- a/gcc/config/riscv/riscv-vector-builtins-shapes.h
> +++ b/gcc/config/riscv/riscv-vector-builtins-shapes.h
> @@ -35,6 +35,7 @@ extern const function_shape *const widen_alu_frm;
> extern const function_shape *const no_mask_policy;
> extern const function_shape *const return_mask;
> extern const function_shape *const narrow_alu;
> +extern const function_shape *const narrow_alu_frm;
> extern const function_shape *const move;
> extern const function_shape *const mask_alu;
> extern const function_shape *const reduc_alu;
> diff --git a/gcc/testsuite/gcc.target/riscv/rvv/base/float-point-ncvt-x.c b/gcc/testsuite/gcc.target/riscv/rvv/base/float-point-ncvt-x.c
> new file mode 100644
> index 00000000000..1630b7e7ccf
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/base/float-point-ncvt-x.c
> @@ -0,0 +1,29 @@
> +/* { dg-do compile } */
> +/* { dg-options "-march=rv64gcv -mabi=lp64 -O3 -Wno-psabi" } */
> +
> +#include "riscv_vector.h"
> +
> +vint16mf2_t
> +test_riscv_vfncvt_x_f_w_i16mf2_rm (vfloat32m1_t op1, size_t vl) {
> + return __riscv_vfncvt_x_f_w_i16mf2_rm (op1, 0, vl);
> +}
> +
> +vint16mf2_t
> +test_vfncvt_x_f_w_i16mf2_rm_m (vbool32_t mask, vfloat32m1_t op1, size_t vl) {
> + return __riscv_vfncvt_x_f_w_i16mf2_rm_m (mask, op1, 1, vl);
> +}
> +
> +vint16mf2_t
> +test_riscv_vfncvt_x_f_w_i16mf2 (vfloat32m1_t op1, size_t vl) {
> + return __riscv_vfncvt_x_f_w_i16mf2 (op1, vl);
> +}
> +
> +vint16mf2_t
> +test_vfncvt_x_f_w_i16mf2_m (vbool32_t mask, vfloat32m1_t op1, size_t vl) {
> + return __riscv_vfncvt_x_f_w_i16mf2_m (mask, op1, vl);
> +}
> +
> +/* { dg-final { scan-assembler-times {vfncvt\.x\.f\.w\s+v[0-9]+,\s*v[0-9]+} 4 } } */
> +/* { dg-final { scan-assembler-times {frrm\s+[axs][0-9]+} 2 } } */
> +/* { dg-final { scan-assembler-times {fsrm\s+[axs][0-9]+} 2 } } */
> +/* { dg-final { scan-assembler-times {fsrmi\s+[01234]} 2 } } */
> --
> 2.34.1
>
@@ -1759,10 +1759,15 @@ public:
};
/* Implements vfncvt.x. */
-template<int UNSPEC>
+template<int UNSPEC, enum frm_op_type FRM_OP = NO_FRM>
class vfncvt_x : public function_base
{
public:
+ bool has_rounding_mode_operand_p () const override
+ {
+ return FRM_OP == HAS_FRM;
+ }
+
rtx expand (function_expander &e) const override
{
return e.use_exact_insn (
@@ -2502,6 +2507,7 @@ static CONSTEXPR const vfwcvt_rtz_x<FIX> vfwcvt_rtz_x_obj;
static CONSTEXPR const vfwcvt_rtz_x<UNSIGNED_FIX> vfwcvt_rtz_xu_obj;
static CONSTEXPR const vfwcvt_f vfwcvt_f_obj;
static CONSTEXPR const vfncvt_x<UNSPEC_VFCVT> vfncvt_x_obj;
+static CONSTEXPR const vfncvt_x<UNSPEC_VFCVT, HAS_FRM> vfncvt_x_frm_obj;
static CONSTEXPR const vfncvt_x<UNSPEC_UNSIGNED_VFCVT> vfncvt_xu_obj;
static CONSTEXPR const vfncvt_rtz_x<FIX> vfncvt_rtz_x_obj;
static CONSTEXPR const vfncvt_rtz_x<UNSIGNED_FIX> vfncvt_rtz_xu_obj;
@@ -2756,6 +2762,7 @@ BASE (vfwcvt_rtz_x)
BASE (vfwcvt_rtz_xu)
BASE (vfwcvt_f)
BASE (vfncvt_x)
+BASE (vfncvt_x_frm)
BASE (vfncvt_xu)
BASE (vfncvt_rtz_x)
BASE (vfncvt_rtz_xu)
@@ -220,6 +220,7 @@ extern const function_base *const vfwcvt_rtz_x;
extern const function_base *const vfwcvt_rtz_xu;
extern const function_base *const vfwcvt_f;
extern const function_base *const vfncvt_x;
+extern const function_base *const vfncvt_x_frm;
extern const function_base *const vfncvt_xu;
extern const function_base *const vfncvt_rtz_x;
extern const function_base *const vfncvt_rtz_xu;
@@ -472,6 +472,8 @@ DEF_RVV_FUNCTION (vfncvt_f, narrow_alu, full_preds, u_to_nf_xu_w_ops)
DEF_RVV_FUNCTION (vfncvt_f, narrow_alu, full_preds, f_to_nf_f_w_ops)
DEF_RVV_FUNCTION (vfncvt_rod_f, narrow_alu, full_preds, f_to_nf_f_w_ops)
+DEF_RVV_FUNCTION (vfncvt_x_frm, narrow_alu_frm, full_preds, f_to_ni_f_w_ops)
+
/* 14. Vector Reduction Operations. */
// 14.1. Vector Single-Width Integer Reduction Instructions
@@ -333,6 +333,44 @@ struct widen_alu_frm_def : public build_frm_base
}
};
+/* narrow_alu_frm_def class. */
+struct narrow_alu_frm_def : public build_frm_base
+{
+ char *get_name (function_builder &b, const function_instance &instance,
+ bool overloaded_p) const override
+ {
+ char base_name[BASE_NAME_MAX_LEN] = {};
+
+ normalize_base_name (base_name, instance.base_name, sizeof (base_name));
+
+ b.append_base_name (base_name);
+
+ if (!overloaded_p)
+ {
+ /* vop --> vop_<op>. */
+ b.append_name (operand_suffixes[instance.op_info->op]);
+ /* vop_<op> --> vop_<op>_<type>. */
+ vector_type_index ret_type_idx
+ = instance.op_info->ret.get_function_type_index (instance.type.index);
+ b.append_name (type_suffixes[ret_type_idx].vector);
+ }
+
+ /* According to rvv-intrinsic-doc, it does not add "_rm" suffix
+ for vop_rm C++ overloaded API. */
+ if (!overloaded_p)
+ b.append_name ("_rm");
+
+ /* According to rvv-intrinsic-doc, it does not add "_m" suffix
+ for vop_m C++ overloaded API. */
+ if (overloaded_p && instance.pred == PRED_TYPE_m)
+ return b.finish_name ();
+
+ b.append_name (predication_suffixes[instance.pred]);
+
+ return b.finish_name ();
+ }
+};
+
/* widen_alu_def class. Handle vwadd/vwsub. Unlike
vadd.vx/vadd.vv/vwmul.vv/vwmul.vx, vwadd.vv/vwadd.vx/vwadd.wv/vwadd.wx has
'OP' suffix in overloaded API. */
@@ -856,6 +894,7 @@ SHAPE(widen_alu_frm, widen_alu_frm)
SHAPE(no_mask_policy, no_mask_policy)
SHAPE(return_mask, return_mask)
SHAPE(narrow_alu, narrow_alu)
+SHAPE(narrow_alu_frm, narrow_alu_frm)
SHAPE(move, move)
SHAPE(mask_alu, mask_alu)
SHAPE(reduc_alu, reduc_alu)
@@ -35,6 +35,7 @@ extern const function_shape *const widen_alu_frm;
extern const function_shape *const no_mask_policy;
extern const function_shape *const return_mask;
extern const function_shape *const narrow_alu;
+extern const function_shape *const narrow_alu_frm;
extern const function_shape *const move;
extern const function_shape *const mask_alu;
extern const function_shape *const reduc_alu;
new file mode 100644
@@ -0,0 +1,29 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64 -O3 -Wno-psabi" } */
+
+#include "riscv_vector.h"
+
+vint16mf2_t
+test_riscv_vfncvt_x_f_w_i16mf2_rm (vfloat32m1_t op1, size_t vl) {
+ return __riscv_vfncvt_x_f_w_i16mf2_rm (op1, 0, vl);
+}
+
+vint16mf2_t
+test_vfncvt_x_f_w_i16mf2_rm_m (vbool32_t mask, vfloat32m1_t op1, size_t vl) {
+ return __riscv_vfncvt_x_f_w_i16mf2_rm_m (mask, op1, 1, vl);
+}
+
+vint16mf2_t
+test_riscv_vfncvt_x_f_w_i16mf2 (vfloat32m1_t op1, size_t vl) {
+ return __riscv_vfncvt_x_f_w_i16mf2 (op1, vl);
+}
+
+vint16mf2_t
+test_vfncvt_x_f_w_i16mf2_m (vbool32_t mask, vfloat32m1_t op1, size_t vl) {
+ return __riscv_vfncvt_x_f_w_i16mf2_m (mask, op1, vl);
+}
+
+/* { dg-final { scan-assembler-times {vfncvt\.x\.f\.w\s+v[0-9]+,\s*v[0-9]+} 4 } } */
+/* { dg-final { scan-assembler-times {frrm\s+[axs][0-9]+} 2 } } */
+/* { dg-final { scan-assembler-times {fsrm\s+[axs][0-9]+} 2 } } */
+/* { dg-final { scan-assembler-times {fsrmi\s+[01234]} 2 } } */