diff mbox series

[V2] RISC-V: Support COND_LEN_* patterns

Message ID 20230712152438.296209-1-juzhe.zhong@rivai.ai
State New
Headers show
Series [V2] RISC-V: Support COND_LEN_* patterns | expand

Commit Message

juzhe.zhong@rivai.ai July 12, 2023, 3:24 p.m. UTC 
This middle-end has been merged:
https://github.com/gcc-mirror/gcc/commit/0d4dd7e07a879d6c07a33edb2799710faa95651e

With this patch, we can handle operations may trap on elements outside the loop.
 
These 2 following cases will be addressed by this patch:
 
1. integer division:
 
  #define TEST_TYPE(TYPE) \
  __attribute__((noipa)) \
  void vrem_##TYPE (TYPE * __restrict dst, TYPE * __restrict a, TYPE * __restrict b, int n) \
  { \
    for (int i = 0; i < n; i++) \
      dst[i] = a[i] % b[i]; \
  }
  #define TEST_ALL() \
   TEST_TYPE(int8_t) \
  TEST_ALL()
 
  Before this patch:
 
   vrem_int8_t:
        ble     a3,zero,.L14
        csrr    t4,vlenb
        addiw   a5,a3,-1
        addiw   a4,t4,-1
        sext.w  t5,a3
        bltu    a5,a4,.L10
        csrr    t3,vlenb
        subw    t3,t5,t3
        li      a5,0
        vsetvli t6,zero,e8,m1,ta,ma
.L4:
        add     a6,a2,a5
        add     a7,a0,a5
        add     t1,a1,a5
        mv      a4,a5
        add     a5,a5,t4
        vl1re8.v        v2,0(a6)
        vl1re8.v        v1,0(t1)
        sext.w  a6,a5
        vrem.vv v1,v1,v2
        vs1r.v  v1,0(a7)
        bleu    a6,t3,.L4
        csrr    a5,vlenb
        addw    a4,a4,a5
        sext.w  a5,a4
        beq     t5,a4,.L16
.L3:
        csrr    a6,vlenb
        subw    t5,t5,a4
        srli    a6,a6,1
        addiw   t1,t5,-1
        addiw   a7,a6,-1
        bltu    t1,a7,.L9
        slli    a4,a4,32
        srli    a4,a4,32
        add     t0,a1,a4
        add     t6,a2,a4
        add     a4,a0,a4
        vsetvli a7,zero,e8,mf2,ta,ma
        sext.w  t3,a6
        vle8.v  v1,0(t0)
        vle8.v  v2,0(t6)
        subw    t4,t5,a6
        vrem.vv v1,v1,v2
        vse8.v  v1,0(a4)
        mv      t1,t3
        bltu    t4,t3,.L7
        csrr    t1,vlenb
        add     a4,a4,a6
        add     t0,t0,a6
        add     t6,t6,a6
        sext.w  t1,t1
        vle8.v  v1,0(t0)
        vle8.v  v2,0(t6)
        vrem.vv v1,v1,v2
        vse8.v  v1,0(a4)
.L7:
        addw    a5,t1,a5
        beq     t5,t1,.L14
.L9:
        add     a4,a1,a5
        add     a6,a2,a5
        lb      a6,0(a6)
        lb      a4,0(a4)
        add     a7,a0,a5
        addi    a5,a5,1
        remw    a4,a4,a6
        sext.w  a6,a5
        sb      a4,0(a7)
        bgt     a3,a6,.L9
.L14:
        ret
.L10:
        li      a4,0
        li      a5,0
        j       .L3
.L16:
        ret
 
After this patch:
 
   vrem_int8_t:
ble a3,zero,.L5
.L3:
vsetvli a5,a3,e8,m1,tu,ma
vle8.v v1,0(a1)
vle8.v v2,0(a2)
sub a3,a3,a5
vrem.vv v1,v1,v2
vse8.v v1,0(a0)
add a1,a1,a5
add a2,a2,a5
add a0,a0,a5
bne a3,zero,.L3
.L5:
ret
 
2. Floating-point operation **WITHOUT** -ffast-math:
     
    #define TEST_TYPE(TYPE) \
    __attribute__((noipa)) \
    void vadd_##TYPE (TYPE * __restrict dst, TYPE *__restrict a, TYPE *__restrict b, int n) \
    { \
      for (int i = 0; i < n; i++) \
        dst[i] = a[i] + b[i]; \
    }
 
    #define TEST_ALL() \
     TEST_TYPE(float) \
 
    TEST_ALL()
   
Before this patch:
   
   vadd_float:
        ble     a3,zero,.L10
        csrr    a4,vlenb
        srli    t3,a4,2
        addiw   a5,a3,-1
        addiw   a6,t3,-1
        sext.w  t6,a3
        bltu    a5,a6,.L7
        subw    t5,t6,t3
        mv      t1,a1
        mv      a7,a2
        mv      a6,a0
        li      a5,0
        vsetvli t4,zero,e32,m1,ta,ma
.L4:
        vl1re32.v       v1,0(t1)
        vl1re32.v       v2,0(a7)
        addw    a5,a5,t3
        vfadd.vv        v1,v1,v2
        vs1r.v  v1,0(a6)
        add     t1,t1,a4
        add     a7,a7,a4
        add     a6,a6,a4
        bgeu    t5,a5,.L4
        beq     t6,a5,.L10
        sext.w  a5,a5
.L3:
        slli    a4,a5,2
.L6:
        add     a6,a1,a4
        add     a7,a2,a4
        flw     fa4,0(a6)
        flw     fa5,0(a7)
        add     a6,a0,a4
        addiw   a5,a5,1
        fadd.s  fa5,fa5,fa4
        addi    a4,a4,4
        fsw     fa5,0(a6)
        bgt     a3,a5,.L6
.L10:
        ret
.L7:
        li      a5,0
        j       .L3
 
After this patch:
 
   vadd_float:
ble a3,zero,.L5
.L3:
vsetvli a5,a3,e32,m1,tu,ma
slli a4,a5,2
vle32.v v1,0(a1)
vle32.v v2,0(a2)
sub a3,a3,a5
vfadd.vv v1,v1,v2
vse32.v v1,0(a0)
add a1,a1,a4
add a2,a2,a4
add a0,a0,a4
bne a3,zero,.L3
.L5:
ret
  
gcc/ChangeLog:
 
        * config/riscv/autovec.md (cond_len_<optab><mode>): New pattern.
        * config/riscv/riscv-protos.h (enum insn_type): New enum.
        (expand_cond_len_binop): New function.
        * config/riscv/riscv-v.cc (emit_nonvlmax_tu_insn): Ditto.
        (emit_nonvlmax_fp_tu_insn): Ditto.
        (need_fp_rounding_p): Ditto.
        (expand_cond_len_binop): Ditto.
        * config/riscv/riscv.cc (riscv_preferred_else_value): Ditto.
        (TARGET_PREFERRED_ELSE_VALUE): New target hook.
 
gcc/testsuite/ChangeLog:
 
        * gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv.c: Adapt testcase.
        * gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv.c: Ditto.
        * gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c: Ditto.
        * gcc.target/riscv/rvv/autovec/binop/vrem-rv64gcv.c: Ditto.
        * gcc.target/riscv/rvv/autovec/binop/vadd-run-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vadd-rv32gcv-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vadd-rv64gcv-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vdiv-run-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vmul-run-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vsub-run-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vsub-rv32gcv-nofm.c: New test.
        * gcc.target/riscv/rvv/autovec/binop/vsub-rv64gcv-nofm.c: New test.

---
 gcc/config/riscv/autovec.md                   | 67 ++++++++++++++++
 gcc/config/riscv/riscv-protos.h               |  2 +
 gcc/config/riscv/riscv-v.cc                   | 79 +++++++++++++++++++
 gcc/config/riscv/riscv.cc                     | 21 +++++
 .../riscv/rvv/autovec/binop/vadd-run-nofm.c   |  4 +
 .../rvv/autovec/binop/vadd-rv32gcv-nofm.c     | 13 +++
 .../rvv/autovec/binop/vadd-rv64gcv-nofm.c     | 13 +++
 .../riscv/rvv/autovec/binop/vdiv-run-nofm.c   |  4 +
 .../rvv/autovec/binop/vdiv-rv32gcv-nofm.c     | 12 +++
 .../riscv/rvv/autovec/binop/vdiv-rv32gcv.c    | 11 ++-
 .../rvv/autovec/binop/vdiv-rv64gcv-nofm.c     | 12 +++
 .../riscv/rvv/autovec/binop/vdiv-rv64gcv.c    | 11 ++-
 .../riscv/rvv/autovec/binop/vmul-run-nofm.c   |  4 +
 .../rvv/autovec/binop/vmul-rv32gcv-nofm.c     |  8 ++
 .../rvv/autovec/binop/vmul-rv64gcv-nofm.c     |  8 ++
 .../riscv/rvv/autovec/binop/vrem-rv32gcv.c    | 10 +--
 .../riscv/rvv/autovec/binop/vrem-rv64gcv.c    | 10 +--
 .../riscv/rvv/autovec/binop/vsub-run-nofm.c   |  4 +
 .../rvv/autovec/binop/vsub-rv32gcv-nofm.c     | 14 ++++
 .../rvv/autovec/binop/vsub-rv64gcv-nofm.c     | 14 ++++
 20 files changed, 297 insertions(+), 24 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-run-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv32gcv-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv64gcv-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-run-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-run-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-run-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv32gcv-nofm.c
 create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv64gcv-nofm.c

Comments

Jeff Law July 12, 2023, 3:40 p.m. UTC | #1 
On 7/12/23 09:24, Juzhe-Zhong wrote:
> This middle-end has been merged:
> https://github.com/gcc-mirror/gcc/commit/0d4dd7e07a879d6c07a33edb2799710faa95651e
> 
> With this patch, we can handle operations may trap on elements outside the loop.
>   
> These 2 following cases will be addressed by this patch:
>   
> 1. integer division:
>   
>    #define TEST_TYPE(TYPE) \
>    __attribute__((noipa)) \
>    void vrem_##TYPE (TYPE * __restrict dst, TYPE * __restrict a, TYPE * __restrict b, int n) \
>    { \
>      for (int i = 0; i < n; i++) \
>        dst[i] = a[i] % b[i]; \
>    }
>    #define TEST_ALL() \
>     TEST_TYPE(int8_t) \
>    TEST_ALL()
>   
>    Before this patch:
>   
>     vrem_int8_t:
>          ble     a3,zero,.L14
>          csrr    t4,vlenb
>          addiw   a5,a3,-1
>          addiw   a4,t4,-1
>          sext.w  t5,a3
>          bltu    a5,a4,.L10
>          csrr    t3,vlenb
>          subw    t3,t5,t3
>          li      a5,0
>          vsetvli t6,zero,e8,m1,ta,ma
> .L4:
>          add     a6,a2,a5
>          add     a7,a0,a5
>          add     t1,a1,a5
>          mv      a4,a5
>          add     a5,a5,t4
>          vl1re8.v        v2,0(a6)
>          vl1re8.v        v1,0(t1)
>          sext.w  a6,a5
>          vrem.vv v1,v1,v2
>          vs1r.v  v1,0(a7)
>          bleu    a6,t3,.L4
>          csrr    a5,vlenb
>          addw    a4,a4,a5
>          sext.w  a5,a4
>          beq     t5,a4,.L16
> .L3:
>          csrr    a6,vlenb
>          subw    t5,t5,a4
>          srli    a6,a6,1
>          addiw   t1,t5,-1
>          addiw   a7,a6,-1
>          bltu    t1,a7,.L9
>          slli    a4,a4,32
>          srli    a4,a4,32
>          add     t0,a1,a4
>          add     t6,a2,a4
>          add     a4,a0,a4
>          vsetvli a7,zero,e8,mf2,ta,ma
>          sext.w  t3,a6
>          vle8.v  v1,0(t0)
>          vle8.v  v2,0(t6)
>          subw    t4,t5,a6
>          vrem.vv v1,v1,v2
>          vse8.v  v1,0(a4)
>          mv      t1,t3
>          bltu    t4,t3,.L7
>          csrr    t1,vlenb
>          add     a4,a4,a6
>          add     t0,t0,a6
>          add     t6,t6,a6
>          sext.w  t1,t1
>          vle8.v  v1,0(t0)
>          vle8.v  v2,0(t6)
>          vrem.vv v1,v1,v2
>          vse8.v  v1,0(a4)
> .L7:
>          addw    a5,t1,a5
>          beq     t5,t1,.L14
> .L9:
>          add     a4,a1,a5
>          add     a6,a2,a5
>          lb      a6,0(a6)
>          lb      a4,0(a4)
>          add     a7,a0,a5
>          addi    a5,a5,1
>          remw    a4,a4,a6
>          sext.w  a6,a5
>          sb      a4,0(a7)
>          bgt     a3,a6,.L9
> .L14:
>          ret
> .L10:
>          li      a4,0
>          li      a5,0
>          j       .L3
> .L16:
>          ret
>   
> After this patch:
>   
>     vrem_int8_t:
> ble a3,zero,.L5
> .L3:
> vsetvli a5,a3,e8,m1,tu,ma
> vle8.v v1,0(a1)
> vle8.v v2,0(a2)
> sub a3,a3,a5
> vrem.vv v1,v1,v2
> vse8.v v1,0(a0)
> add a1,a1,a5
> add a2,a2,a5
> add a0,a0,a5
> bne a3,zero,.L3
> .L5:
> ret
>   
> 2. Floating-point operation **WITHOUT** -ffast-math:
>       
>      #define TEST_TYPE(TYPE) \
>      __attribute__((noipa)) \
>      void vadd_##TYPE (TYPE * __restrict dst, TYPE *__restrict a, TYPE *__restrict b, int n) \
>      { \
>        for (int i = 0; i < n; i++) \
>          dst[i] = a[i] + b[i]; \
>      }
>   
>      #define TEST_ALL() \
>       TEST_TYPE(float) \
>   
>      TEST_ALL()
>     
> Before this patch:
>     
>     vadd_float:
>          ble     a3,zero,.L10
>          csrr    a4,vlenb
>          srli    t3,a4,2
>          addiw   a5,a3,-1
>          addiw   a6,t3,-1
>          sext.w  t6,a3
>          bltu    a5,a6,.L7
>          subw    t5,t6,t3
>          mv      t1,a1
>          mv      a7,a2
>          mv      a6,a0
>          li      a5,0
>          vsetvli t4,zero,e32,m1,ta,ma
> .L4:
>          vl1re32.v       v1,0(t1)
>          vl1re32.v       v2,0(a7)
>          addw    a5,a5,t3
>          vfadd.vv        v1,v1,v2
>          vs1r.v  v1,0(a6)
>          add     t1,t1,a4
>          add     a7,a7,a4
>          add     a6,a6,a4
>          bgeu    t5,a5,.L4
>          beq     t6,a5,.L10
>          sext.w  a5,a5
> .L3:
>          slli    a4,a5,2
> .L6:
>          add     a6,a1,a4
>          add     a7,a2,a4
>          flw     fa4,0(a6)
>          flw     fa5,0(a7)
>          add     a6,a0,a4
>          addiw   a5,a5,1
>          fadd.s  fa5,fa5,fa4
>          addi    a4,a4,4
>          fsw     fa5,0(a6)
>          bgt     a3,a5,.L6
> .L10:
>          ret
> .L7:
>          li      a5,0
>          j       .L3
>   
> After this patch:
>   
>     vadd_float:
> ble a3,zero,.L5
> .L3:
> vsetvli a5,a3,e32,m1,tu,ma
> slli a4,a5,2
> vle32.v v1,0(a1)
> vle32.v v2,0(a2)
> sub a3,a3,a5
> vfadd.vv v1,v1,v2
> vse32.v v1,0(a0)
> add a1,a1,a4
> add a2,a2,a4
> add a0,a0,a4
> bne a3,zero,.L3
> .L5:
> ret
>    
> gcc/ChangeLog:
>   
>          * config/riscv/autovec.md (cond_len_<optab><mode>): New pattern.
>          * config/riscv/riscv-protos.h (enum insn_type): New enum.
>          (expand_cond_len_binop): New function.
>          * config/riscv/riscv-v.cc (emit_nonvlmax_tu_insn): Ditto.
>          (emit_nonvlmax_fp_tu_insn): Ditto.
>          (need_fp_rounding_p): Ditto.
>          (expand_cond_len_binop): Ditto.
>          * config/riscv/riscv.cc (riscv_preferred_else_value): Ditto.
>          (TARGET_PREFERRED_ELSE_VALUE): New target hook.
>   
> gcc/testsuite/ChangeLog:
>   
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv.c: Adapt testcase.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv.c: Ditto.
>          * gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c: Ditto.
>          * gcc.target/riscv/rvv/autovec/binop/vrem-rv64gcv.c: Ditto.
>          * gcc.target/riscv/rvv/autovec/binop/vadd-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vadd-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vadd-rv64gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vmul-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vsub-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vsub-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vsub-rv64gcv-nofm.c: New test.
OK with the change to call needs_fp_rounding and the minor comment fix 
for FMLA/FMACC are addressed.  No need to wait for another review round. 
  Just post and commit it as pre-approved.

jeff
Li, Pan2 via Gcc-patches July 12, 2023, 11:48 p.m. UTC | #2 
Committed, thanks Jeff.

Pan

-----Original Message-----
From: Gcc-patches <gcc-patches-bounces+pan2.li=intel.com@gcc.gnu.org> On Behalf Of Jeff Law via Gcc-patches
Sent: Wednesday, July 12, 2023 11:40 PM
To: Juzhe-Zhong <juzhe.zhong@rivai.ai>; gcc-patches@gcc.gnu.org
Cc: kito.cheng@sifive.com; kito.cheng@gmail.com; rdapp.gcc@gmail.com
Subject: Re: [PATCH V2] RISC-V: Support COND_LEN_* patterns



On 7/12/23 09:24, Juzhe-Zhong wrote:
> This middle-end has been merged:
> https://github.com/gcc-mirror/gcc/commit/0d4dd7e07a879d6c07a33edb2799710faa95651e
> 
> With this patch, we can handle operations may trap on elements outside the loop.
>   
> These 2 following cases will be addressed by this patch:
>   
> 1. integer division:
>   
>    #define TEST_TYPE(TYPE) \
>    __attribute__((noipa)) \
>    void vrem_##TYPE (TYPE * __restrict dst, TYPE * __restrict a, TYPE * __restrict b, int n) \
>    { \
>      for (int i = 0; i < n; i++) \
>        dst[i] = a[i] % b[i]; \
>    }
>    #define TEST_ALL() \
>     TEST_TYPE(int8_t) \
>    TEST_ALL()
>   
>    Before this patch:
>   
>     vrem_int8_t:
>          ble     a3,zero,.L14
>          csrr    t4,vlenb
>          addiw   a5,a3,-1
>          addiw   a4,t4,-1
>          sext.w  t5,a3
>          bltu    a5,a4,.L10
>          csrr    t3,vlenb
>          subw    t3,t5,t3
>          li      a5,0
>          vsetvli t6,zero,e8,m1,ta,ma
> .L4:
>          add     a6,a2,a5
>          add     a7,a0,a5
>          add     t1,a1,a5
>          mv      a4,a5
>          add     a5,a5,t4
>          vl1re8.v        v2,0(a6)
>          vl1re8.v        v1,0(t1)
>          sext.w  a6,a5
>          vrem.vv v1,v1,v2
>          vs1r.v  v1,0(a7)
>          bleu    a6,t3,.L4
>          csrr    a5,vlenb
>          addw    a4,a4,a5
>          sext.w  a5,a4
>          beq     t5,a4,.L16
> .L3:
>          csrr    a6,vlenb
>          subw    t5,t5,a4
>          srli    a6,a6,1
>          addiw   t1,t5,-1
>          addiw   a7,a6,-1
>          bltu    t1,a7,.L9
>          slli    a4,a4,32
>          srli    a4,a4,32
>          add     t0,a1,a4
>          add     t6,a2,a4
>          add     a4,a0,a4
>          vsetvli a7,zero,e8,mf2,ta,ma
>          sext.w  t3,a6
>          vle8.v  v1,0(t0)
>          vle8.v  v2,0(t6)
>          subw    t4,t5,a6
>          vrem.vv v1,v1,v2
>          vse8.v  v1,0(a4)
>          mv      t1,t3
>          bltu    t4,t3,.L7
>          csrr    t1,vlenb
>          add     a4,a4,a6
>          add     t0,t0,a6
>          add     t6,t6,a6
>          sext.w  t1,t1
>          vle8.v  v1,0(t0)
>          vle8.v  v2,0(t6)
>          vrem.vv v1,v1,v2
>          vse8.v  v1,0(a4)
> .L7:
>          addw    a5,t1,a5
>          beq     t5,t1,.L14
> .L9:
>          add     a4,a1,a5
>          add     a6,a2,a5
>          lb      a6,0(a6)
>          lb      a4,0(a4)
>          add     a7,a0,a5
>          addi    a5,a5,1
>          remw    a4,a4,a6
>          sext.w  a6,a5
>          sb      a4,0(a7)
>          bgt     a3,a6,.L9
> .L14:
>          ret
> .L10:
>          li      a4,0
>          li      a5,0
>          j       .L3
> .L16:
>          ret
>   
> After this patch:
>   
>     vrem_int8_t:
> ble a3,zero,.L5
> .L3:
> vsetvli a5,a3,e8,m1,tu,ma
> vle8.v v1,0(a1)
> vle8.v v2,0(a2)
> sub a3,a3,a5
> vrem.vv v1,v1,v2
> vse8.v v1,0(a0)
> add a1,a1,a5
> add a2,a2,a5
> add a0,a0,a5
> bne a3,zero,.L3
> .L5:
> ret
>   
> 2. Floating-point operation **WITHOUT** -ffast-math:
>       
>      #define TEST_TYPE(TYPE) \
>      __attribute__((noipa)) \
>      void vadd_##TYPE (TYPE * __restrict dst, TYPE *__restrict a, TYPE *__restrict b, int n) \
>      { \
>        for (int i = 0; i < n; i++) \
>          dst[i] = a[i] + b[i]; \
>      }
>   
>      #define TEST_ALL() \
>       TEST_TYPE(float) \
>   
>      TEST_ALL()
>     
> Before this patch:
>     
>     vadd_float:
>          ble     a3,zero,.L10
>          csrr    a4,vlenb
>          srli    t3,a4,2
>          addiw   a5,a3,-1
>          addiw   a6,t3,-1
>          sext.w  t6,a3
>          bltu    a5,a6,.L7
>          subw    t5,t6,t3
>          mv      t1,a1
>          mv      a7,a2
>          mv      a6,a0
>          li      a5,0
>          vsetvli t4,zero,e32,m1,ta,ma
> .L4:
>          vl1re32.v       v1,0(t1)
>          vl1re32.v       v2,0(a7)
>          addw    a5,a5,t3
>          vfadd.vv        v1,v1,v2
>          vs1r.v  v1,0(a6)
>          add     t1,t1,a4
>          add     a7,a7,a4
>          add     a6,a6,a4
>          bgeu    t5,a5,.L4
>          beq     t6,a5,.L10
>          sext.w  a5,a5
> .L3:
>          slli    a4,a5,2
> .L6:
>          add     a6,a1,a4
>          add     a7,a2,a4
>          flw     fa4,0(a6)
>          flw     fa5,0(a7)
>          add     a6,a0,a4
>          addiw   a5,a5,1
>          fadd.s  fa5,fa5,fa4
>          addi    a4,a4,4
>          fsw     fa5,0(a6)
>          bgt     a3,a5,.L6
> .L10:
>          ret
> .L7:
>          li      a5,0
>          j       .L3
>   
> After this patch:
>   
>     vadd_float:
> ble a3,zero,.L5
> .L3:
> vsetvli a5,a3,e32,m1,tu,ma
> slli a4,a5,2
> vle32.v v1,0(a1)
> vle32.v v2,0(a2)
> sub a3,a3,a5
> vfadd.vv v1,v1,v2
> vse32.v v1,0(a0)
> add a1,a1,a4
> add a2,a2,a4
> add a0,a0,a4
> bne a3,zero,.L3
> .L5:
> ret
>    
> gcc/ChangeLog:
>   
>          * config/riscv/autovec.md (cond_len_<optab><mode>): New pattern.
>          * config/riscv/riscv-protos.h (enum insn_type): New enum.
>          (expand_cond_len_binop): New function.
>          * config/riscv/riscv-v.cc (emit_nonvlmax_tu_insn): Ditto.
>          (emit_nonvlmax_fp_tu_insn): Ditto.
>          (need_fp_rounding_p): Ditto.
>          (expand_cond_len_binop): Ditto.
>          * config/riscv/riscv.cc (riscv_preferred_else_value): Ditto.
>          (TARGET_PREFERRED_ELSE_VALUE): New target hook.
>   
> gcc/testsuite/ChangeLog:
>   
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv.c: Adapt testcase.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv.c: Ditto.
>          * gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c: Ditto.
>          * gcc.target/riscv/rvv/autovec/binop/vrem-rv64gcv.c: Ditto.
>          * gcc.target/riscv/rvv/autovec/binop/vadd-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vadd-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vadd-rv64gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vmul-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vsub-run-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vsub-rv32gcv-nofm.c: New test.
>          * gcc.target/riscv/rvv/autovec/binop/vsub-rv64gcv-nofm.c: New test.
OK with the change to call needs_fp_rounding and the minor comment fix 
for FMLA/FMACC are addressed.  No need to wait for another review round. 
  Just post and commit it as pre-approved.

jeff
diff mbox series

Patch

diff --git a/gcc/config/riscv/autovec.md b/gcc/config/riscv/autovec.md
index d98a63c285e..883ac4747fd 100644
--- a/gcc/config/riscv/autovec.md
+++ b/gcc/config/riscv/autovec.md
@@ -1208,3 +1208,70 @@ 
   riscv_vector::emit_vlmax_insn (icode, riscv_vector::RVV_BINOP, operands);
   DONE;
 })
+
+;; -------------------------------------------------------------------------
+;; ---- [INT] Conditional binary operations
+;; -------------------------------------------------------------------------
+;; Includes:
+;; - vadd.vv/vsub.vv/...
+;; - vadd.vi/vsub.vi/...
+;; -------------------------------------------------------------------------
+
+(define_expand "cond_len_<optab><mode>"
+  [(match_operand:VI 0 "register_operand")
+   (match_operand:<VM> 1 "vector_mask_operand")
+   (any_int_binop_no_shift:VI
+     (match_operand:VI 2 "<binop_rhs1_predicate>")
+     (match_operand:VI 3 "<binop_rhs2_predicate>"))
+   (match_operand:VI 4 "register_operand")
+   (match_operand 5 "autovec_length_operand")
+   (match_operand 6 "const_0_operand")]
+  "TARGET_VECTOR"
+{
+  riscv_vector::expand_cond_len_binop (<CODE>, operands);
+  DONE;
+})
+
+;; -------------------------------------------------------------------------
+;; ---- [FP] Conditional binary operations
+;; -------------------------------------------------------------------------
+;; Includes:
+;; - vfadd.vv/vfsub.vv/...
+;; - vfadd.vf/vfsub.vf/...
+;; -------------------------------------------------------------------------
+
+(define_expand "cond_len_<optab><mode>"
+  [(match_operand:VF 0 "register_operand")
+   (match_operand:<VM> 1 "vector_mask_operand")
+   (any_float_binop:VF
+     (match_operand:VF 2 "register_operand")
+     (match_operand:VF 3 "register_operand"))
+   (match_operand:VF 4 "register_operand")
+   (match_operand 5 "autovec_length_operand")
+   (match_operand 6 "const_0_operand")]
+  "TARGET_VECTOR"
+{
+  riscv_vector::expand_cond_len_binop (<CODE>, operands);
+  DONE;
+})
+
+;; -------------------------------------------------------------------------
+;; Includes:
+;; - vfmin.vv/vfmax.vv
+;; - vfmin.vf/vfmax.vf
+;; -------------------------------------------------------------------------
+
+(define_expand "cond_len_<optab><mode>"
+  [(match_operand:VF 0 "register_operand")
+   (match_operand:<VM> 1 "vector_mask_operand")
+   (any_float_binop_nofrm:VF
+     (match_operand:VF 2 "register_operand")
+     (match_operand:VF 3 "register_operand"))
+   (match_operand:VF 4 "register_operand")
+   (match_operand 5 "autovec_length_operand")
+   (match_operand 6 "const_0_operand")]
+  "TARGET_VECTOR"
+{
+  riscv_vector::expand_cond_len_binop (<CODE>, operands);
+  DONE;
+})
diff --git a/gcc/config/riscv/riscv-protos.h b/gcc/config/riscv/riscv-protos.h
index 5ce0d470272..b7b53958871 100644
--- a/gcc/config/riscv/riscv-protos.h
+++ b/gcc/config/riscv/riscv-protos.h
@@ -184,6 +184,7 @@  enum insn_type
   RVV_UNOP = 2,
   RVV_BINOP = 3,
   RVV_BINOP_MU = RVV_BINOP + 2,
+  RVV_BINOP_TU = RVV_BINOP + 2,
   RVV_MERGE_OP = 4,
   RVV_CMP_OP = 4,
   RVV_CMP_MU_OP = RVV_CMP_OP + 2, /* +2 means mask and maskoff operand.  */
@@ -276,6 +277,7 @@  bool neg_simm5_p (rtx);
 bool has_vi_variant_p (rtx_code, rtx);
 void expand_vec_cmp (rtx, rtx_code, rtx, rtx);
 bool expand_vec_cmp_float (rtx, rtx_code, rtx, rtx, bool);
+void expand_cond_len_binop (rtx_code, rtx *);
 #endif
 bool sew64_scalar_helper (rtx *, rtx *, rtx, machine_mode,
 			  bool, void (*)(rtx *, rtx));
diff --git a/gcc/config/riscv/riscv-v.cc b/gcc/config/riscv/riscv-v.cc
index 1a61f9a18cd..9d3f588ce89 100644
--- a/gcc/config/riscv/riscv-v.cc
+++ b/gcc/config/riscv/riscv-v.cc
@@ -919,6 +919,45 @@  emit_vlmax_masked_mu_insn (unsigned icode, int op_num, rtx *ops)
   e.emit_insn ((enum insn_code) icode, ops);
 }
 
+/* This function emits a TU instruction.  */
+static void
+emit_nonvlmax_tu_insn (unsigned icode, int op_num, rtx *ops, rtx avl)
+{
+  machine_mode dest_mode = GET_MODE (ops[0]);
+  machine_mode mask_mode = get_mask_mode (dest_mode).require ();
+  insn_expander<RVV_INSN_OPERANDS_MAX> e (/*OP_NUM*/ op_num,
+					  /*HAS_DEST_P*/ true,
+					  /*FULLY_UNMASKED_P*/ false,
+					  /*USE_REAL_MERGE_P*/ true,
+					  /*HAS_AVL_P*/ true,
+					  /*VLMAX_P*/ false, dest_mode,
+					  mask_mode);
+  e.set_policy (TAIL_UNDISTURBED);
+  e.set_policy (MASK_ANY);
+  e.set_vl (avl);
+  e.emit_insn ((enum insn_code) icode, ops);
+}
+
+/* This function emits a TU instruction.  */
+static void
+emit_nonvlmax_fp_tu_insn (unsigned icode, int op_num, rtx *ops, rtx avl)
+{
+  machine_mode dest_mode = GET_MODE (ops[0]);
+  machine_mode mask_mode = get_mask_mode (dest_mode).require ();
+  insn_expander<RVV_INSN_OPERANDS_MAX> e (/*OP_NUM*/ op_num,
+					  /*HAS_DEST_P*/ true,
+					  /*FULLY_UNMASKED_P*/ false,
+					  /*USE_REAL_MERGE_P*/ true,
+					  /*HAS_AVL_P*/ true,
+					  /*VLMAX_P*/ false, dest_mode,
+					  mask_mode);
+  e.set_policy (TAIL_UNDISTURBED);
+  e.set_policy (MASK_ANY);
+  e.set_rounding_mode (FRM_DYN);
+  e.set_vl (avl);
+  e.emit_insn ((enum insn_code) icode, ops);
+}
+
 /* Emit vmv.s.x instruction.  */
 
 void
@@ -2968,4 +3007,44 @@  expand_load_store (rtx *ops, bool is_load)
     }
 }
 
+/* Return true if the operation is the floating-point operation need FRM.  */
+static bool
+needs_fp_rounding (rtx_code code, machine_mode mode)
+{
+  if (!FLOAT_MODE_P (mode))
+    return false;
+  return code != SMIN && code != SMAX;
+}
+
+/* Expand COND_LEN_*.  */
+void
+expand_cond_len_binop (rtx_code code, rtx *ops)
+{
+  rtx dest = ops[0];
+  rtx mask = ops[1];
+  rtx src1 = ops[2];
+  rtx src2 = ops[3];
+  rtx merge = ops[4];
+  rtx len = ops[5];
+  machine_mode mode = GET_MODE (dest);
+  machine_mode mask_mode = GET_MODE (mask);
+
+  poly_uint64 value;
+  bool is_dummy_mask = rtx_equal_p (mask, CONSTM1_RTX (mask_mode));
+
+  if (is_dummy_mask)
+    {
+      /* Use TU, MASK ANY policy.  */
+      rtx ops[] = {dest, mask, merge, src1, src2};
+      insn_code icode = code_for_pred (code, mode);
+      if (needs_fp_rounding (code, mode))
+	emit_nonvlmax_fp_tu_insn (icode, RVV_BINOP_MU, ops, len);
+      else
+	emit_nonvlmax_tu_insn (icode, RVV_BINOP_MU, ops, len);
+    }
+  else
+    /* FIXME: Enable this case when we support it in the middle-end.  */
+    gcc_unreachable ();
+}
+
 } // namespace riscv_vector
diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
index af18231437d..398194fe902 100644
--- a/gcc/config/riscv/riscv.cc
+++ b/gcc/config/riscv/riscv.cc
@@ -7832,6 +7832,24 @@  riscv_vectorize_vec_perm_const (machine_mode vmode, machine_mode op_mode,
   return false;
 }
 
+/* Implement TARGET_PREFERRED_ELSE_VALUE.  For binary operations,
+   prefer to use the first arithmetic operand as the else value if
+   the else value doesn't matter, since that exactly matches the RVV
+   destructive merging form.  For ternary operations we could either
+   pick the first operand and use VMADD-like instructions or the last
+   operand and use VMACC-like instructions; the latter seems more
+   natural.
+
+   TODO: Currently, the return value is not ideal for RVV since it will
+   let VSETVL PASS use MU or TU. We will suport undefine value that allows
+   VSETVL PASS use TA/MA in the future.  */
+
+static tree
+riscv_preferred_else_value (unsigned, tree, unsigned int nops, tree *ops)
+{
+  return nops == 3 ? ops[2] : ops[0];
+}
+
 /* Initialize the GCC target structure.  */
 #undef TARGET_ASM_ALIGNED_HI_OP
 #define TARGET_ASM_ALIGNED_HI_OP "\t.half\t"
@@ -8133,6 +8151,9 @@  riscv_vectorize_vec_perm_const (machine_mode vmode, machine_mode op_mode,
 #undef TARGET_VECTORIZE_VEC_PERM_CONST
 #define TARGET_VECTORIZE_VEC_PERM_CONST riscv_vectorize_vec_perm_const
 
+#undef TARGET_PREFERRED_ELSE_VALUE
+#define TARGET_PREFERRED_ELSE_VALUE riscv_preferred_else_value
+
 struct gcc_target targetm = TARGET_INITIALIZER;
 
 #include "gt-riscv.h"
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-run-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-run-nofm.c
new file mode 100644
index 00000000000..66ae1167128
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-run-nofm.c
@@ -0,0 +1,4 @@ 
+/* { dg-do run { target { riscv_vector } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=scalable" } */
+
+#include "vadd-run.c"
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv32gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv32gcv-nofm.c
new file mode 100644
index 00000000000..069bc690697
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv32gcv-nofm.c
@@ -0,0 +1,13 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vadd-template.h"
+
+/* { dg-final { scan-assembler-times {\tvadd\.vv} 16 } } */
+/* { dg-final { scan-assembler-times {\tvadd\.vi} 8 } } */
+/* { dg-final { scan-assembler-times {\tvfadd\.vv} 7 } } */
+/* There are 2 MINUS operations.  */
+/* { dg-final { scan-assembler-times {\tvfsub\.vv} 2 } } */
+
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_ADD" 7 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_SUB" 2 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv64gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv64gcv-nofm.c
new file mode 100644
index 00000000000..07fa54878cc
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vadd-rv64gcv-nofm.c
@@ -0,0 +1,13 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vadd-template.h"
+
+/* { dg-final { scan-assembler-times {\tvadd\.vv} 16 } } */
+/* { dg-final { scan-assembler-times {\tvadd\.vi} 8 } } */
+/* { dg-final { scan-assembler-times {\tvfadd\.vv} 7 } } */
+/* There are 2 MINUS operations.  */
+/* { dg-final { scan-assembler-times {\tvfsub\.vv} 2 } } */
+
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_ADD" 7 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_SUB" 2 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-run-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-run-nofm.c
new file mode 100644
index 00000000000..ed340b840d1
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-run-nofm.c
@@ -0,0 +1,4 @@ 
+/* { dg-do run { target { riscv_vector } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=scalable" } */
+
+#include "vdiv-run.c"
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv-nofm.c
new file mode 100644
index 00000000000..5ce2e57e265
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv-nofm.c
@@ -0,0 +1,12 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vdiv-template.h"
+
+/* { dg-final { scan-assembler-times {\tvdiv\.vv} 8 } } */
+/* { dg-final { scan-assembler-times {\tvdivu\.vv} 8 } } */
+
+/* { dg-final { scan-assembler-times {\tvfdiv\.vv} 6 } } */
+
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_DIV" 16 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_RDIV" 6 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv.c
index 604d9acb038..9b984dda452 100644
--- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv.c
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv32gcv.c
@@ -1,15 +1,14 @@ 
 /* { dg-do compile } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax -ffast-math -fdump-tree-optimized-details" } */
 
 #include "vdiv-template.h"
 
-/* Currently we use an epilogue loop which also contains vdivs.  Therefore we
-   expect 14 vdiv[u]s instead of 8.  */
-
-/* { dg-final { scan-assembler-times {\tvdiv\.vv} 14 } } */
-/* { dg-final { scan-assembler-times {\tvdivu\.vv} 14 } } */
+/* { dg-final { scan-assembler-times {\tvdiv\.vv} 8 } } */
+/* { dg-final { scan-assembler-times {\tvdivu\.vv} 8 } } */
 
 /* Division by constant is done by calculating a reciprocal and
    then multiplying.  Hence we do not expect 6 vfdivs.  */
 /* { dg-final { scan-assembler-times {\tvfdiv\.vv} 3 } } */
 /* { dg-final { scan-assembler-times {\tvfmul\.vv} 3 } } */
+
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_DIV" 16 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv-nofm.c
new file mode 100644
index 00000000000..7b1aa28e45e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv-nofm.c
@@ -0,0 +1,12 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vdiv-template.h"
+
+/* { dg-final { scan-assembler-times {\tvdiv\.vv} 8 } } */
+/* { dg-final { scan-assembler-times {\tvdivu\.vv} 8 } } */
+
+/* { dg-final { scan-assembler-times {\tvfdiv\.vv} 6 } } */
+
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_DIV" 16 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_RDIV" 6 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv.c
index 26884035d57..ca4d23bb1ed 100644
--- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv.c
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vdiv-rv64gcv.c
@@ -1,15 +1,14 @@ 
 /* { dg-do compile } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -ffast-math -fdump-tree-optimized-details" } */
 
 #include "vdiv-template.h"
 
-/* Currently we use an epilogue loop which also contains vdivs.  Therefore we
-   expect 14 vdiv[u]s instead of 8.  */
-
-/* { dg-final { scan-assembler-times {\tvdiv\.vv} 14 } } */
-/* { dg-final { scan-assembler-times {\tvdivu\.vv} 14 } } */
+/* { dg-final { scan-assembler-times {\tvdiv\.vv} 8 } } */
+/* { dg-final { scan-assembler-times {\tvdivu\.vv} 8 } } */
 
 /* Division by constant is done by calculating a reciprocal and
    then multiplying.  Hence we do not expect 6 vfdivs.  */
 /* { dg-final { scan-assembler-times {\tvfdiv\.vv} 3 } } */
 /* { dg-final { scan-assembler-times {\tvfmul\.vv} 3 } } */
+
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_DIV" 16 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-run-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-run-nofm.c
new file mode 100644
index 00000000000..225030e2df0
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-run-nofm.c
@@ -0,0 +1,4 @@ 
+/* { dg-do run { target { riscv_vector } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=scalable" } */
+
+#include "vmul-run.c"
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv-nofm.c
new file mode 100644
index 00000000000..3e0f06162fc
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv-nofm.c
@@ -0,0 +1,8 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vmul-template.h"
+
+/* { dg-final { scan-assembler-times {\tvmul\.vv} 16 } } */
+/* { dg-final { scan-assembler-times {\tvfmul\.vv} 6 } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_MUL" 6 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv-nofm.c
new file mode 100644
index 00000000000..ca245e28662
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv-nofm.c
@@ -0,0 +1,8 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vmul-template.h"
+
+/* { dg-final { scan-assembler-times {\tvmul\.vv} 16 } } */
+/* { dg-final { scan-assembler-times {\tvfmul\.vv} 6 } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_MUL" 6 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c
index 86607d03777..6d0493b3194 100644
--- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c
@@ -1,9 +1,7 @@ 
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax -fdump-tree-optimized-details" } */
 
 #include "vrem-template.h"
 
-/* Currently we use an epilogue loop which also contains vrems.  Therefore we
-   expect 14 vrem[u]s instead of 8.  */
-
-/* { dg-final { scan-assembler-times {\tvrem\.vv} 14 } } */
-/* { dg-final { scan-assembler-times {\tvremu\.vv} 14 } } */
+/* { dg-final { scan-assembler-times {\tvrem\.vv} 8 } } */
+/* { dg-final { scan-assembler-times {\tvremu\.vv} 8 } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_MOD" 16 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv64gcv.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv64gcv.c
index b7bc1ccb860..24b2bc81a90 100644
--- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv64gcv.c
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vrem-rv64gcv.c
@@ -1,10 +1,8 @@ 
 /* { dg-do compile } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -fdump-tree-optimized-details" } */
 
 #include "vrem-template.h"
 
-/* Currently we use an epilogue loop which also contains vrems.  Therefore we
-   expect 14 vrem[u]s instead of 8.  */
-
-/* { dg-final { scan-assembler-times {\tvrem\.vv} 14 } } */
-/* { dg-final { scan-assembler-times {\tvremu\.vv} 14 } } */
+/* { dg-final { scan-assembler-times {\tvrem\.vv} 8 } } */
+/* { dg-final { scan-assembler-times {\tvremu\.vv} 8 } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_MOD" 16 "optimized" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-run-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-run-nofm.c
new file mode 100644
index 00000000000..36a17068e0e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-run-nofm.c
@@ -0,0 +1,4 @@ 
+/* { dg-do run { target { riscv_vector } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=scalable" } */
+
+#include "vsub-run.c"
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv32gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv32gcv-nofm.c
new file mode 100644
index 00000000000..c4ab934cdf5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv32gcv-nofm.c
@@ -0,0 +1,14 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vsub-template.h"
+
+/* { dg-final { scan-assembler-times {\tvsub\.vv} 16 } } */
+/* { dg-final { scan-assembler-times {\tvrsub\.vi} 16 } } */
+
+/* { dg-final { scan-assembler-times {\tvfsub\.vv} 12 } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_SUB" 12 "optimized" } } */
+
+/* Do not expect vfrsub for now, because we do not properly
+   handle vop.vx and vfop.vf yet.  */
+/* { dg-final { scan-assembler-times {\tvfrsub\.vv} 0 } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv64gcv-nofm.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv64gcv-nofm.c
new file mode 100644
index 00000000000..9e71911a92a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/vsub-rv64gcv-nofm.c
@@ -0,0 +1,14 @@ 
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=scalable -fdump-tree-optimized-details" } */
+
+#include "vsub-template.h"
+
+/* { dg-final { scan-assembler-times {\tvsub\.vv} 16 } } */
+/* { dg-final { scan-assembler-times {\tvrsub\.vi} 16 } } */
+
+/* { dg-final { scan-assembler-times {\tvfsub\.vv} 12 } } */
+/* { dg-final { scan-tree-dump-times "\.COND_LEN_SUB" 12 "optimized" } } */
+
+/* Do not expect vfrsub for now, because we do not properly
+   handle vop.vx and vfop.vf yet.  */
+/* { dg-final { scan-assembler-times {\tvfrsub\.vv} 0 } } */