Re: [PATCH 11/38] target/riscv: SIMD 8-bit Multiply Instructions

[Alistair Francis]

On Fri, Feb 12, 2021 at 10:26 AM LIU Zhiwei <zhiwei_liu@c-sky.com> wrote:
>
> Signed-off-by: LIU Zhiwei <zhiwei_liu@c-sky.com>

Acked-by: Alistair Francis <alistair.francis@wdc.com>

Alistair

> ---
>  target/riscv/helper.h                   |  7 ++
>  target/riscv/insn32.decode              |  7 ++
>  target/riscv/insn_trans/trans_rvp.c.inc |  8 +++
>  target/riscv/packed_helper.c            | 93 +++++++++++++++++++++++++
>  4 files changed, 115 insertions(+)
>
> diff --git a/target/riscv/helper.h b/target/riscv/helper.h
> index bc60712bd9..6bb601b436 100644
> --- a/target/riscv/helper.h
> +++ b/target/riscv/helper.h
> @@ -1221,3 +1221,10 @@ DEF_HELPER_3(umul16, i64, env, tl, tl)
>  DEF_HELPER_3(umulx16, i64, env, tl, tl)
>  DEF_HELPER_3(khm16, tl, env, tl, tl)
>  DEF_HELPER_3(khmx16, tl, env, tl, tl)
> +
> +DEF_HELPER_3(smul8, i64, env, tl, tl)
> +DEF_HELPER_3(smulx8, i64, env, tl, tl)
> +DEF_HELPER_3(umul8, i64, env, tl, tl)
> +DEF_HELPER_3(umulx8, i64, env, tl, tl)
> +DEF_HELPER_3(khm8, tl, env, tl, tl)
> +DEF_HELPER_3(khmx8, tl, env, tl, tl)
> diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode
> index 38519a477c..9d165efba9 100644
> --- a/target/riscv/insn32.decode
> +++ b/target/riscv/insn32.decode
> @@ -688,3 +688,10 @@ umul16     1011000  ..... ..... 000 ..... 1111111 @r
>  umulx16    1011001  ..... ..... 000 ..... 1111111 @r
>  khm16      1000011  ..... ..... 000 ..... 1111111 @r
>  khmx16     1001011  ..... ..... 000 ..... 1111111 @r
> +
> +smul8      1010100  ..... ..... 000 ..... 1111111 @r
> +smulx8     1010101  ..... ..... 000 ..... 1111111 @r
> +umul8      1011100  ..... ..... 000 ..... 1111111 @r
> +umulx8     1011101  ..... ..... 000 ..... 1111111 @r
> +khm8       1000111  ..... ..... 000 ..... 1111111 @r
> +khmx8      1001111  ..... ..... 000 ..... 1111111 @r
> diff --git a/target/riscv/insn_trans/trans_rvp.c.inc 
> b/target/riscv/insn_trans/trans_rvp.c.inc
> index 7e5bf9041d..336f3418b1 100644
> --- a/target/riscv/insn_trans/trans_rvp.c.inc
> +++ b/target/riscv/insn_trans/trans_rvp.c.inc
> @@ -436,3 +436,11 @@ GEN_RVP_R_D64_OOL(umul16);
>  GEN_RVP_R_D64_OOL(umulx16);
>  GEN_RVP_R_OOL(khm16);
>  GEN_RVP_R_OOL(khmx16);
> +
> +/* SIMD 8-bit Multiply Instructions */
> +GEN_RVP_R_D64_OOL(smul8);
> +GEN_RVP_R_D64_OOL(smulx8);
> +GEN_RVP_R_D64_OOL(umul8);
> +GEN_RVP_R_D64_OOL(umulx8);
> +GEN_RVP_R_OOL(khm8);
> +GEN_RVP_R_OOL(khmx8);
> diff --git a/target/riscv/packed_helper.c b/target/riscv/packed_helper.c
> index 13fed2c4d1..56baefeb8e 100644
> --- a/target/riscv/packed_helper.c
> +++ b/target/riscv/packed_helper.c
> @@ -827,3 +827,96 @@ static inline void do_khmx16(CPURISCVState *env, void 
> *vd, void *va,
>  }
>
>  RVPR(khmx16, 2, 2);
> +
> +/* SIMD 8-bit Multiply Instructions */
> +static inline void do_smul8(CPURISCVState *env, void *vd, void *va, void *vb)
> +{
> +    int16_t *d = vd;
> +    int8_t *a = va, *b = vb;
> +    d[H2(0)] = (int16_t)a[H1(0)] * b[H1(0)];
> +    d[H2(1)] = (int16_t)a[H1(1)] * b[H1(1)];
> +    d[H2(2)] = (int16_t)a[H1(2)] * b[H1(2)];
> +    d[H2(3)] = (int16_t)a[H1(3)] * b[H1(3)];
> +}
> +
> +RVPR64(smul8);
> +
> +static inline void do_smulx8(CPURISCVState *env, void *vd, void *va, void 
> *vb)
> +{
> +    int16_t *d = vd;
> +    int8_t *a = va, *b = vb;
> +    d[H2(0)] = (int16_t)a[H1(0)] * b[H1(1)];
> +    d[H2(1)] = (int16_t)a[H1(1)] * b[H1(0)];
> +    d[H2(2)] = (int16_t)a[H1(2)] * b[H1(3)];
> +    d[H2(3)] = (int16_t)a[H1(3)] * b[H1(2)];
> +}
> +
> +RVPR64(smulx8);
> +
> +static inline void do_umul8(CPURISCVState *env, void *vd, void *va, void *vb)
> +{
> +    uint16_t *d = vd;
> +    uint8_t *a = va, *b = vb;
> +    d[H2(0)] = (uint16_t)a[H1(0)] * b[H1(0)];
> +    d[H2(1)] = (uint16_t)a[H1(1)] * b[H1(1)];
> +    d[H2(2)] = (uint16_t)a[H1(2)] * b[H1(2)];
> +    d[H2(3)] = (uint16_t)a[H1(3)] * b[H1(3)];
> +}
> +
> +RVPR64(umul8);
> +
> +static inline void do_umulx8(CPURISCVState *env, void *vd, void *va, void 
> *vb)
> +{
> +    uint16_t *d = vd;
> +    uint8_t *a = va, *b = vb;
> +    d[H2(0)] = (uint16_t)a[H1(0)] * b[H1(1)];
> +    d[H2(1)] = (uint16_t)a[H1(1)] * b[H1(0)];
> +    d[H2(2)] = (uint16_t)a[H1(2)] * b[H1(3)];
> +    d[H2(3)] = (uint16_t)a[H1(3)] * b[H1(2)];
> +}
> +
> +RVPR64(umulx8);
> +
> +static inline void do_khm8(CPURISCVState *env, void *vd, void *va,
> +                           void *vb, uint8_t i)
> +{
> +    int8_t *d = vd, *a = va, *b = vb;
> +
> +    if (a[i] == INT8_MIN && b[i] == INT8_MIN) {
> +        env->vxsat = 1;
> +        d[i] = INT8_MAX;
> +    } else {
> +        d[i] = (int16_t)a[i] * b[i] >> 7;
> +    }
> +}
> +
> +RVPR(khm8, 1, 1);
> +
> +static inline void do_khmx8(CPURISCVState *env, void *vd, void *va,
> +                            void *vb, uint8_t i)
> +{
> +    int8_t *d = vd, *a = va, *b = vb;
> +    /*
> +     * t[x] = ra.B[x] s* rb.B[y];
> +     * rt.B[x] = SAT.Q7(t[x] s>> 7);
> +     *
> +     * (RV32: (x,y)=(3,2),(2,3),
> +     *              (1,0),(0,1),
> +     * (RV64: (x,y)=(7,6),(6,7),(5,4),(4,5),
> +     *              (3,2),(2,3),(1,0),(0,1))
> +     */
> +    if (a[H1(i)] == INT8_MIN && b[H1(i + 1)] == INT8_MIN) {
> +        env->vxsat = 1;
> +        d[H1(i)] = INT8_MAX;
> +    } else {
> +        d[H1(i)] = (int16_t)a[H1(i)] * b[H1(i + 1)] >> 7;
> +    }
> +    if (a[H1(i + 1)] == INT8_MIN && b[H1(i)] == INT8_MIN) {
> +        env->vxsat = 1;
> +        d[H1(i + 1)] = INT8_MAX;
> +    } else {
> +        d[H1(i + 1)] = (int16_t)a[H1(i + 1)] * b[H1(i)] >> 7;
> +    }
> +}
> +
> +RVPR(khmx8, 2, 1);
> --
> 2.17.1
>