Re: [PATCH] RISC-V: Fix pred_mov constraint for vle.v

2023-01-27 Thread Kito Cheng via Gcc-patches
Committed, thanks!

On Thu, Jan 19, 2023 at 3:03 PM  wrote:

> From: Ju-Zhe Zhong 
>
> The original constraint is incorrect in pred_mov pattern.
> Take a look at Alternative 2, the operands[0] is "vr",
> operands[1] which is mask operand can be "vm".
> Such alternative matching will give the wrong codegen (vle.v v0,0(a5),v0.t)
> This is illegal according to RVV ISA.
>
> To fix this issue and not destroy the RA performance, fix this pattern in
> this patch.
>
> gcc/ChangeLog:
>
> * config/riscv/vector.md: Fix constraints.
>
> ---
>  gcc/config/riscv/vector.md | 29 +++--
>  1 file changed, 15 insertions(+), 14 deletions(-)
>
> diff --git a/gcc/config/riscv/vector.md b/gcc/config/riscv/vector.md
> index 48414e200cf..e1173f2d5a6 100644
> --- a/gcc/config/riscv/vector.md
> +++ b/gcc/config/riscv/vector.md
> @@ -633,22 +633,23 @@
>  ;;2. (const_vector:VNx1SF repeat [
>  ;;(const_double:SF 0.0 [0x0.0p+0])]).
>  (define_insn_and_split "@pred_mov"
> -  [(set (match_operand:V 0 "nonimmediate_operand"  "=vd,vr,
>m,vr,vr")
> -   (if_then_else:V
> - (unspec:
> -   [(match_operand: 1 "vector_mask_operand" "vmWc1, vmWc1,
> vmWc1,   Wc1,   Wc1")
> -(match_operand 4 "vector_length_operand""   rK,rK,
> rK,rK,rK")
> -(match_operand 5 "const_int_operand""i, i,
>  i, i, i")
> -(match_operand 6 "const_int_operand""i, i,
>  i, i, i")
> -(match_operand 7 "const_int_operand""i, i,
>  i, i, i")
> -(reg:SI VL_REGNUM)
> -(reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
> - (match_operand:V 3 "vector_move_operand"   "m, m,
> vr,vr, viWc0")
> - (match_operand:V 2 "vector_merge_operand"  "0,vu,
> vu,   vu0,   vu0")))]
> +  [(set (match_operand:V 0 "nonimmediate_operand"  "=vr,vr,
> vd, m,vr,vr")
> +(if_then_else:V
> +  (unspec:
> +[(match_operand: 1 "vector_mask_operand" "vmWc1,   Wc1,
> vm, vmWc1,   Wc1,   Wc1")
> + (match_operand 4 "vector_length_operand""   rK,rK,
> rK,rK,rK,rK")
> + (match_operand 5 "const_int_operand""i, i,
>  i, i, i, i")
> + (match_operand 6 "const_int_operand""i, i,
>  i, i, i, i")
> + (match_operand 7 "const_int_operand""i, i,
>  i, i, i, i")
> + (reg:SI VL_REGNUM)
> + (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
> +  (match_operand:V 3 "vector_move_operand"   "m, m,
>  m,vr,vr, viWc0")
> +  (match_operand:V 2 "vector_merge_operand"  "0,vu,
> vu,vu,   vu0,   vu0")))]
>"TARGET_VECTOR"
>"@
> vle.v\t%0,%3%p1
> -   vle.v\t%0,%3%p1
> +   vle.v\t%0,%3
> +   vle.v\t%0,%3,%1.t
> vse.v\t%3,%0%p1
> vmv.v.v\t%0,%3
> vmv.v.i\t%0,%v3"
> @@ -657,7 +658,7 @@
> && satisfies_constraint_vu (operands[2])"
>[(set (match_dup 0) (match_dup 3))]
>""
> -  [(set_attr "type" "vlde,vlde,vste,vimov,vimov")
> +  [(set_attr "type" "vlde,vlde,vlde,vste,vimov,vimov")
> (set_attr "mode" "")])
>
>  ;; Dedicated pattern for vse.v instruction since we can't reuse pred_mov
> pattern to include
> --
> 2.36.3
>
>


[PATCH] RISC-V: Fix pred_mov constraint for vle.v

2023-01-18 Thread juzhe . zhong
From: Ju-Zhe Zhong 

The original constraint is incorrect in pred_mov pattern.
Take a look at Alternative 2, the operands[0] is "vr",
operands[1] which is mask operand can be "vm".
Such alternative matching will give the wrong codegen (vle.v v0,0(a5),v0.t)
This is illegal according to RVV ISA.

To fix this issue and not destroy the RA performance, fix this pattern in
this patch.
 
gcc/ChangeLog:

* config/riscv/vector.md: Fix constraints.

---
 gcc/config/riscv/vector.md | 29 +++--
 1 file changed, 15 insertions(+), 14 deletions(-)

diff --git a/gcc/config/riscv/vector.md b/gcc/config/riscv/vector.md
index 48414e200cf..e1173f2d5a6 100644
--- a/gcc/config/riscv/vector.md
+++ b/gcc/config/riscv/vector.md
@@ -633,22 +633,23 @@
 ;;2. (const_vector:VNx1SF repeat [
 ;;(const_double:SF 0.0 [0x0.0p+0])]).
 (define_insn_and_split "@pred_mov"
-  [(set (match_operand:V 0 "nonimmediate_operand"  "=vd,vr, m, 
   vr,vr")
-   (if_then_else:V
- (unspec:
-   [(match_operand: 1 "vector_mask_operand" "vmWc1, vmWc1, vmWc1,  
 Wc1,   Wc1")
-(match_operand 4 "vector_length_operand""   rK,rK,rK,  
  rK,rK")
-(match_operand 5 "const_int_operand""i, i, i,  
   i, i")
-(match_operand 6 "const_int_operand""i, i, i,  
   i, i")
-(match_operand 7 "const_int_operand""i, i, i,  
   i, i")
-(reg:SI VL_REGNUM)
-(reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
- (match_operand:V 3 "vector_move_operand"   "m, m,vr,  
  vr, viWc0")
- (match_operand:V 2 "vector_merge_operand"  "0,vu,vu,  
 vu0,   vu0")))]
+  [(set (match_operand:V 0 "nonimmediate_operand"  "=vr,vr,vd, 
m,vr,vr")
+(if_then_else:V
+  (unspec:
+[(match_operand: 1 "vector_mask_operand" "vmWc1,   Wc1,vm, 
vmWc1,   Wc1,   Wc1")
+ (match_operand 4 "vector_length_operand""   rK,rK,rK,
rK,rK,rK")
+ (match_operand 5 "const_int_operand""i, i, i, 
i, i, i")
+ (match_operand 6 "const_int_operand""i, i, i, 
i, i, i")
+ (match_operand 7 "const_int_operand""i, i, i, 
i, i, i")
+ (reg:SI VL_REGNUM)
+ (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
+  (match_operand:V 3 "vector_move_operand"   "m, m, m,
vr,vr, viWc0")
+  (match_operand:V 2 "vector_merge_operand"  "0,vu,vu,
vu,   vu0,   vu0")))]
   "TARGET_VECTOR"
   "@
vle.v\t%0,%3%p1
-   vle.v\t%0,%3%p1
+   vle.v\t%0,%3
+   vle.v\t%0,%3,%1.t
vse.v\t%3,%0%p1
vmv.v.v\t%0,%3
vmv.v.i\t%0,%v3"
@@ -657,7 +658,7 @@
&& satisfies_constraint_vu (operands[2])"
   [(set (match_dup 0) (match_dup 3))]
   ""
-  [(set_attr "type" "vlde,vlde,vste,vimov,vimov")
+  [(set_attr "type" "vlde,vlde,vlde,vste,vimov,vimov")
(set_attr "mode" "")])
 
 ;; Dedicated pattern for vse.v instruction since we can't reuse pred_mov 
pattern to include
-- 
2.36.3