Re: [Qemu-riscv] [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> wrote:

From: LIU Zhiwei <address@hidden>

Signed-off-by: LIU Zhiwei <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