Re: [Qemu-devel] [PATCH v2 01/17] RISC-V: add vfp field in CPURISCVState

[liuzhiwei]

On 2019/9/11 下午10:51, Chih-Min Chao wrote:
> On Wed, Sep 11, 2019 at 2:35 PM liuzhiwei <address@hidden 
> <mailto:address@hidden>> wrote:
>
>     From: LIU Zhiwei <address@hidden <mailto:address@hidden>>
>
>     Signed-off-by: LIU Zhiwei <address@hidden
>     <mailto:address@hidden>>
>     ---
>      target/riscv/cpu.h | 28 ++++++++++++++++++++++++++++
>      1 file changed, 28 insertions(+)
>
>     diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
>     index 0adb307..c992b1d 100644
>     --- a/target/riscv/cpu.h
>     +++ b/target/riscv/cpu.h
>     @@ -93,9 +93,37 @@ typedef struct CPURISCVState CPURISCVState;
>
>      #include "pmp.h"
>
>     +#define VLEN 128
>     +#define VUNIT(x) (VLEN / x)
>     +
>      struct CPURISCVState {
>          target_ulong gpr[32];
>          uint64_t fpr[32]; /* assume both F and D extensions */
>     +
>     +    /* vector coprocessor state.  */
>     +    struct {
>     +        union VECTOR {
>     +            float64  f64[VUNIT(64)];
>     +            float32  f32[VUNIT(32)];
>     +            float16  f16[VUNIT(16)];
>     +            uint64_t u64[VUNIT(64)];
>     +            int64_t  s64[VUNIT(64)];
>     +            uint32_t u32[VUNIT(32)];
>     +            int32_t  s32[VUNIT(32)];
>     +            uint16_t u16[VUNIT(16)];
>     +            int16_t  s16[VUNIT(16)];
>     +            uint8_t  u8[VUNIT(8)];
>     +            int8_t   s8[VUNIT(8)];
>     +        } vreg[32];
>     +        target_ulong vxrm;
>     +        target_ulong vxsat;
>     +        target_ulong vl;
>     +        target_ulong vstart;
>     +        target_ulong vtype;
>     +        float_status fp_status;
>     +    } vfp;
>     +
>     +    bool         foflag;
>          target_ulong pc;
>          target_ulong load_res;
>          target_ulong load_val;
>     -- 
>     2.7.4
>
>
> Could  the VLEN be configurable in cpu initialization but not fixed in 
> compilation phase ?

Yes,  it's important that VLEN is configurable to support different 
types of cpu.

> Take the integer element as example  and the difference should be the 
> stride of vfp.vreg[x] isn't continuous
>
>     struct {
>         union VECTOR {
>             uint64_t *u64;
>             uint16_t *u16;
>             uint8_t  *u8;
>         } vreg[32];
>     } vfp;
>
>    initialization
>     int vlen = 256;  //parameter from cpu command line option
>     int elem = vlen / 8;
>     int size = elem * 32;
>
>     uint8_t *mem = malloc(size)
>     for (int idx = 0; idx < 32; ++idx) {
>         vfp.vreg[idx].u64 = (void *)&mem[idx * elem];
>         vfp.vreg[idx].u32 = (void *)&mem[idx * elem];
>         vfp.vreg[idx].u16 = (void *)&mem[idx * elem];
>    }
>
>   chihmin

It's a good idea. I will accept it.

Thanks for review.

Zhiwei