RE: [PATCH v1] RISC-V: Support rounding mode for VFMSAC/VFMSUB autovec

2023-08-31 Thread Li, Pan2 via Gcc-patches
Committed, thanks Kito.

Pan

-Original Message-
From: Kito Cheng  
Sent: Thursday, August 31, 2023 9:09 PM
To: Li, Pan2 
Cc: gcc-patches@gcc.gnu.org; juzhe.zh...@rivai.ai; Wang, Yanzhang 

Subject: Re: [PATCH v1] RISC-V: Support rounding mode for VFMSAC/VFMSUB autovec

LGTM

On Thu, Aug 24, 2023 at 3:13 PM Pan Li via Gcc-patches
 wrote:
>
> From: Pan Li 
>
> There will be a case like below for intrinsic and autovec combination.
>
> vfadd RTZ   <- intrinisc static rounding
> vfmsub  <- autovec/autovec-opt
>
> The autovec generated vfmsub should take DYN mode, and the
> frm must be restored before the vfmsub insn. This patch
> would like to fix this issue by:
>
> * Add the frm operand to the autovec/autovec-opt pattern.
> * Set the frm_mode attr to DYN.
>
> Thus, the frm flow when combine autovec and intrinsic should be.
>
> +
> | frrm  a5
> | ...
> | fsrmi 4
> | vfadd   <- intrinsic static rounding.
> | ...
> | fsrm  a5
> | vfmsub  <- autovec/autovec-opt
> | ...
> +
>
> Signed-off-by: Pan Li 
>
> gcc/ChangeLog:
>
> * config/riscv/autovec-opt.md: Add FRM_REGNUM to vfmsac/vfmsub
> * config/riscv/autovec.md: Ditto.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/riscv/rvv/base/float-point-frm-autovec-2.c: New test.
> ---
>  gcc/config/riscv/autovec-opt.md   | 36 
>  gcc/config/riscv/autovec.md   | 30 ---
>  .../rvv/base/float-point-frm-autovec-2.c  | 88 +++
>  3 files changed, 127 insertions(+), 27 deletions(-)
>  create mode 100644 
> gcc/testsuite/gcc.target/riscv/rvv/base/float-point-frm-autovec-2.c
>
> diff --git a/gcc/config/riscv/autovec-opt.md b/gcc/config/riscv/autovec-opt.md
> index 4b07e80ad95..732a51edacd 100644
> --- a/gcc/config/riscv/autovec-opt.md
> +++ b/gcc/config/riscv/autovec-opt.md
> @@ -583,13 +583,15 @@ (define_insn_and_split "*single_widen_fnma"
>  ;; vect__13.182_33 = .FMS (vect__11.180_35, vect__8.176_40, vect__4.172_45);
>  (define_insn_and_split "*double_widen_fms"
>[(set (match_operand:VWEXTF 0 "register_operand")
> -   (fma:VWEXTF
> - (float_extend:VWEXTF
> -   (match_operand: 2 "register_operand"))
> - (float_extend:VWEXTF
> -   (match_operand: 3 "register_operand"))
> - (neg:VWEXTF
> -   (match_operand:VWEXTF 1 "register_operand"]
> +   (unspec:VWEXTF
> + [(fma:VWEXTF
> +   (float_extend:VWEXTF
> + (match_operand: 2 "register_operand"))
> +   (float_extend:VWEXTF
> + (match_operand: 3 "register_operand"))
> +   (neg:VWEXTF
> + (match_operand:VWEXTF 1 "register_operand")))
> +  (reg:SI FRM_REGNUM)] UNSPEC_VFFMA))]
>"TARGET_VECTOR && can_create_pseudo_p ()"
>"#"
>"&& 1"
> @@ -600,17 +602,20 @@ (define_insn_and_split "*double_widen_fms"
>  DONE;
>}
>[(set_attr "type" "vfwmuladd")
> -   (set_attr "mode" "")])
> +   (set_attr "mode" "")
> +   (set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
>
>  ;; This helps to match ext + fms.
>  (define_insn_and_split "*single_widen_fms"
>[(set (match_operand:VWEXTF 0 "register_operand")
> -   (fma:VWEXTF
> - (float_extend:VWEXTF
> -   (match_operand: 2 "register_operand"))
> - (match_operand:VWEXTF 3 "register_operand")
> - (neg:VWEXTF
> -   (match_operand:VWEXTF 1 "register_operand"]
> +   (unspec:VWEXTF
> + [(fma:VWEXTF
> +   (float_extend:VWEXTF
> + (match_operand: 2 "register_operand"))
> +   (match_operand:VWEXTF 3 "register_operand")
> +   (neg:VWEXTF
> + (match_operand:VWEXTF 1 "register_operand")))
> +  (reg:SI FRM_REGNUM)] UNSPEC_VFFMA))]
>"TARGET_VECTOR && can_create_pseudo_p ()"
>"#"
>"&& 1"
> @@ -627,7 +632,8 @@ (define_insn_and_split "*single_widen_fms"
>  DONE;
>}
>[(set_attr "type" "vfwmuladd")
> -   (set_attr "mode" "")])
> +   (set_attr "mode" "")
> +   (set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
>
>  ;; -
>  ;;  [FP] VFWN

Re: [PATCH v1] RISC-V: Support rounding mode for VFMSAC/VFMSUB autovec

2023-08-31 Thread Kito Cheng via Gcc-patches
LGTM

On Thu, Aug 24, 2023 at 3:13 PM Pan Li via Gcc-patches
 wrote:
>
> From: Pan Li 
>
> There will be a case like below for intrinsic and autovec combination.
>
> vfadd RTZ   <- intrinisc static rounding
> vfmsub  <- autovec/autovec-opt
>
> The autovec generated vfmsub should take DYN mode, and the
> frm must be restored before the vfmsub insn. This patch
> would like to fix this issue by:
>
> * Add the frm operand to the autovec/autovec-opt pattern.
> * Set the frm_mode attr to DYN.
>
> Thus, the frm flow when combine autovec and intrinsic should be.
>
> +
> | frrm  a5
> | ...
> | fsrmi 4
> | vfadd   <- intrinsic static rounding.
> | ...
> | fsrm  a5
> | vfmsub  <- autovec/autovec-opt
> | ...
> +
>
> Signed-off-by: Pan Li 
>
> gcc/ChangeLog:
>
> * config/riscv/autovec-opt.md: Add FRM_REGNUM to vfmsac/vfmsub
> * config/riscv/autovec.md: Ditto.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/riscv/rvv/base/float-point-frm-autovec-2.c: New test.
> ---
>  gcc/config/riscv/autovec-opt.md   | 36 
>  gcc/config/riscv/autovec.md   | 30 ---
>  .../rvv/base/float-point-frm-autovec-2.c  | 88 +++
>  3 files changed, 127 insertions(+), 27 deletions(-)
>  create mode 100644 
> gcc/testsuite/gcc.target/riscv/rvv/base/float-point-frm-autovec-2.c
>
> diff --git a/gcc/config/riscv/autovec-opt.md b/gcc/config/riscv/autovec-opt.md
> index 4b07e80ad95..732a51edacd 100644
> --- a/gcc/config/riscv/autovec-opt.md
> +++ b/gcc/config/riscv/autovec-opt.md
> @@ -583,13 +583,15 @@ (define_insn_and_split "*single_widen_fnma"
>  ;; vect__13.182_33 = .FMS (vect__11.180_35, vect__8.176_40, vect__4.172_45);
>  (define_insn_and_split "*double_widen_fms"
>[(set (match_operand:VWEXTF 0 "register_operand")
> -   (fma:VWEXTF
> - (float_extend:VWEXTF
> -   (match_operand: 2 "register_operand"))
> - (float_extend:VWEXTF
> -   (match_operand: 3 "register_operand"))
> - (neg:VWEXTF
> -   (match_operand:VWEXTF 1 "register_operand"]
> +   (unspec:VWEXTF
> + [(fma:VWEXTF
> +   (float_extend:VWEXTF
> + (match_operand: 2 "register_operand"))
> +   (float_extend:VWEXTF
> + (match_operand: 3 "register_operand"))
> +   (neg:VWEXTF
> + (match_operand:VWEXTF 1 "register_operand")))
> +  (reg:SI FRM_REGNUM)] UNSPEC_VFFMA))]
>"TARGET_VECTOR && can_create_pseudo_p ()"
>"#"
>"&& 1"
> @@ -600,17 +602,20 @@ (define_insn_and_split "*double_widen_fms"
>  DONE;
>}
>[(set_attr "type" "vfwmuladd")
> -   (set_attr "mode" "")])
> +   (set_attr "mode" "")
> +   (set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
>
>  ;; This helps to match ext + fms.
>  (define_insn_and_split "*single_widen_fms"
>[(set (match_operand:VWEXTF 0 "register_operand")
> -   (fma:VWEXTF
> - (float_extend:VWEXTF
> -   (match_operand: 2 "register_operand"))
> - (match_operand:VWEXTF 3 "register_operand")
> - (neg:VWEXTF
> -   (match_operand:VWEXTF 1 "register_operand"]
> +   (unspec:VWEXTF
> + [(fma:VWEXTF
> +   (float_extend:VWEXTF
> + (match_operand: 2 "register_operand"))
> +   (match_operand:VWEXTF 3 "register_operand")
> +   (neg:VWEXTF
> + (match_operand:VWEXTF 1 "register_operand")))
> +  (reg:SI FRM_REGNUM)] UNSPEC_VFFMA))]
>"TARGET_VECTOR && can_create_pseudo_p ()"
>"#"
>"&& 1"
> @@ -627,7 +632,8 @@ (define_insn_and_split "*single_widen_fms"
>  DONE;
>}
>[(set_attr "type" "vfwmuladd")
> -   (set_attr "mode" "")])
> +   (set_attr "mode" "")
> +   (set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
>
>  ;; -
>  ;;  [FP] VFWNMACC
> diff --git a/gcc/config/riscv/autovec.md b/gcc/config/riscv/autovec.md
> index 4894986d2a5..d9f1a10eb66 100644
> --- a/gcc/config/riscv/autovec.md
> +++ b/gcc/config/riscv/autovec.md
> @@ -1218,24 +1218,29 @@ (define_insn_and_split "*fnma"
>  (define_expand "fms4"
>[(parallel
>  [(set (match_operand:VF 0 "register_operand")
> - (fma:VF
> -   (match_operand:VF 1 "register_operand")
> -   (match_operand:VF 2 "register_operand")
> -   (neg:VF
> - (match_operand:VF 3 "register_operand"
> + (unspec:VF
> +   [(fma:VF
> + (match_operand:VF 1 "register_operand")
> + (match_operand:VF 2 "register_operand")
> + (neg:VF
> +   (match_operand:VF 3 "register_operand")))
> +(reg:SI FRM_REGNUM)] UNSPEC_VFFMA))
>   (clobber (match_dup 4))])]
>"TARGET_VECTOR"
>{
>  operands[4] = gen_reg_rtx (Pmode);
> -  })
> +  }
> +  [(set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
>
>  (define_insn_and_split "*fms"
> 

[PATCH v1] RISC-V: Support rounding mode for VFMSAC/VFMSUB autovec

2023-08-24 Thread Pan Li via Gcc-patches
From: Pan Li 

There will be a case like below for intrinsic and autovec combination.

vfadd RTZ   <- intrinisc static rounding
vfmsub  <- autovec/autovec-opt

The autovec generated vfmsub should take DYN mode, and the
frm must be restored before the vfmsub insn. This patch
would like to fix this issue by:

* Add the frm operand to the autovec/autovec-opt pattern.
* Set the frm_mode attr to DYN.

Thus, the frm flow when combine autovec and intrinsic should be.

+
| frrm  a5
| ...
| fsrmi 4
| vfadd   <- intrinsic static rounding.
| ...
| fsrm  a5
| vfmsub  <- autovec/autovec-opt
| ...
+

Signed-off-by: Pan Li 

gcc/ChangeLog:

* config/riscv/autovec-opt.md: Add FRM_REGNUM to vfmsac/vfmsub
* config/riscv/autovec.md: Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/base/float-point-frm-autovec-2.c: New test.
---
 gcc/config/riscv/autovec-opt.md   | 36 
 gcc/config/riscv/autovec.md   | 30 ---
 .../rvv/base/float-point-frm-autovec-2.c  | 88 +++
 3 files changed, 127 insertions(+), 27 deletions(-)
 create mode 100644 
gcc/testsuite/gcc.target/riscv/rvv/base/float-point-frm-autovec-2.c

diff --git a/gcc/config/riscv/autovec-opt.md b/gcc/config/riscv/autovec-opt.md
index 4b07e80ad95..732a51edacd 100644
--- a/gcc/config/riscv/autovec-opt.md
+++ b/gcc/config/riscv/autovec-opt.md
@@ -583,13 +583,15 @@ (define_insn_and_split "*single_widen_fnma"
 ;; vect__13.182_33 = .FMS (vect__11.180_35, vect__8.176_40, vect__4.172_45);
 (define_insn_and_split "*double_widen_fms"
   [(set (match_operand:VWEXTF 0 "register_operand")
-   (fma:VWEXTF
- (float_extend:VWEXTF
-   (match_operand: 2 "register_operand"))
- (float_extend:VWEXTF
-   (match_operand: 3 "register_operand"))
- (neg:VWEXTF
-   (match_operand:VWEXTF 1 "register_operand"]
+   (unspec:VWEXTF
+ [(fma:VWEXTF
+   (float_extend:VWEXTF
+ (match_operand: 2 "register_operand"))
+   (float_extend:VWEXTF
+ (match_operand: 3 "register_operand"))
+   (neg:VWEXTF
+ (match_operand:VWEXTF 1 "register_operand")))
+  (reg:SI FRM_REGNUM)] UNSPEC_VFFMA))]
   "TARGET_VECTOR && can_create_pseudo_p ()"
   "#"
   "&& 1"
@@ -600,17 +602,20 @@ (define_insn_and_split "*double_widen_fms"
 DONE;
   }
   [(set_attr "type" "vfwmuladd")
-   (set_attr "mode" "")])
+   (set_attr "mode" "")
+   (set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
 
 ;; This helps to match ext + fms.
 (define_insn_and_split "*single_widen_fms"
   [(set (match_operand:VWEXTF 0 "register_operand")
-   (fma:VWEXTF
- (float_extend:VWEXTF
-   (match_operand: 2 "register_operand"))
- (match_operand:VWEXTF 3 "register_operand")
- (neg:VWEXTF
-   (match_operand:VWEXTF 1 "register_operand"]
+   (unspec:VWEXTF
+ [(fma:VWEXTF
+   (float_extend:VWEXTF
+ (match_operand: 2 "register_operand"))
+   (match_operand:VWEXTF 3 "register_operand")
+   (neg:VWEXTF
+ (match_operand:VWEXTF 1 "register_operand")))
+  (reg:SI FRM_REGNUM)] UNSPEC_VFFMA))]
   "TARGET_VECTOR && can_create_pseudo_p ()"
   "#"
   "&& 1"
@@ -627,7 +632,8 @@ (define_insn_and_split "*single_widen_fms"
 DONE;
   }
   [(set_attr "type" "vfwmuladd")
-   (set_attr "mode" "")])
+   (set_attr "mode" "")
+   (set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
 
 ;; -
 ;;  [FP] VFWNMACC
diff --git a/gcc/config/riscv/autovec.md b/gcc/config/riscv/autovec.md
index 4894986d2a5..d9f1a10eb66 100644
--- a/gcc/config/riscv/autovec.md
+++ b/gcc/config/riscv/autovec.md
@@ -1218,24 +1218,29 @@ (define_insn_and_split "*fnma"
 (define_expand "fms4"
   [(parallel
 [(set (match_operand:VF 0 "register_operand")
- (fma:VF
-   (match_operand:VF 1 "register_operand")
-   (match_operand:VF 2 "register_operand")
-   (neg:VF
- (match_operand:VF 3 "register_operand"
+ (unspec:VF
+   [(fma:VF
+ (match_operand:VF 1 "register_operand")
+ (match_operand:VF 2 "register_operand")
+ (neg:VF
+   (match_operand:VF 3 "register_operand")))
+(reg:SI FRM_REGNUM)] UNSPEC_VFFMA))
  (clobber (match_dup 4))])]
   "TARGET_VECTOR"
   {
 operands[4] = gen_reg_rtx (Pmode);
-  })
+  }
+  [(set (attr "frm_mode") (symbol_ref "riscv_vector::FRM_DYN"))])
 
 (define_insn_and_split "*fms"
   [(set (match_operand:VF 0 "register_operand" "=vr, vr, ?&vr")
-   (fma:VF
- (match_operand:VF 1 "register_operand"   " %0, vr,   vr")
- (match_operand:VF 2 "register_operand"   " vr, vr,   vr")
- (neg:VF
-   (match_operand:VF 3 "register_operand" " vr,  0,   vr"
+   (unspec:VF
+