Implement the vqwdotau.vv and vqwdotas.vv instructions from the Zvqwdota8i extension. These instructions compute a dot product between two 8-bit integer vectors, accumulating into a 32-bit scalar in vd[0].
Signed-off-by: Max Chou <[email protected]> --- target/riscv/helper.h | 4 ++ target/riscv/insn32.decode | 4 ++ target/riscv/tcg/insn_trans/trans_rvv.c.inc | 63 +++++++++++++++++++++ target/riscv/tcg/vector_helper.c | 49 ++++++++++++++++ 4 files changed, 120 insertions(+) diff --git a/target/riscv/helper.h b/target/riscv/helper.h index 1cffcad580c..767b04658b4 100644 --- a/target/riscv/helper.h +++ b/target/riscv/helper.h @@ -1387,6 +1387,10 @@ DEF_HELPER_1(ssamoswap_disabled, void, env) /* Zalrsc SC write probe */ DEF_HELPER_FLAGS_3(sc_probe_write, TCG_CALL_NO_WG, void, env, tl, tl) +/* Zvqwdota8i - 8-bit Integer Dot Product */ +DEF_HELPER_6(vqwdotau_vv_b, void, ptr, ptr, ptr, ptr, env, i32) +DEF_HELPER_6(vqwdotas_vv_b, void, ptr, ptr, ptr, ptr, env, i32) + /* SiFive Custom int8 Matrix-Multiply */ DEF_HELPER_5(sf_vqmaccu_4x8x4, void, ptr, ptr, ptr, env, i32) DEF_HELPER_5(sf_vqmacc_4x8x4, void, ptr, ptr, ptr, env, i32) diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode index 795badc8a7d..cc7d639ab73 100644 --- a/target/riscv/insn32.decode +++ b/target/riscv/insn32.decode @@ -1097,3 +1097,7 @@ vwabda_vv 111101 . ..... ..... 000 ..... 1010111 @r_vm vwabda_vx 111101 . ..... ..... 100 ..... 1010111 @r_vm vwabdau_vv 111110 . ..... ..... 000 ..... 1010111 @r_vm vwabdau_vx 111110 . ..... ..... 100 ..... 1010111 @r_vm + +# *** Zvqwdota8i 8-bit Integer Dot-Product Extension *** +vqwdotau_vv 100110 . ..... ..... 000 ..... 1110111 @r_vm +vqwdotas_vv 100111 . ..... ..... 000 ..... 1110111 @r_vm diff --git a/target/riscv/tcg/insn_trans/trans_rvv.c.inc b/target/riscv/tcg/insn_trans/trans_rvv.c.inc index 7e1295f33e7..befdbf273e5 100644 --- a/target/riscv/tcg/insn_trans/trans_rvv.c.inc +++ b/target/riscv/tcg/insn_trans/trans_rvv.c.inc @@ -4132,7 +4132,70 @@ static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ return false; \ } +/* Zvqwdota8i 8-bit Integer Dot-Product Extension */ +static bool vext_zvdota_base_check(DisasContext *s, arg_rmrr * a) +{ + return require_rvv(s) && + vext_check_isa_ill(s) && + s->vstart_eq_zero && + require_align(a->rs1, s->lmul) && + require_align(a->rs2, s->lmul) && + require_vm(a->vm, a->rd) && + require_noover(a->rd, 0, a->rs1, s->lmul) && + require_noover(a->rd, 0, a->rs2, s->lmul) && + vext_check_input_eew(s, a->rs1, s->sew, a->rs2, s->sew, a->vm); +} + +static bool vext_vqwdota_check(DisasContext *s, arg_rmrr *a) +{ + return vext_zvdota_base_check(s, a) && + (s->cfg_ptr->ext_zvqwdota8i && s->sew == MO_8); +} + +typedef void gen_helper_zvdota_int(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_ptr, + TCGv_env, TCGv_i32); + +#define GEN_ZVDOTA_INT_TRANS(NAME, CHECK) \ +static bool trans_##NAME(DisasContext *s, arg_rmrr *a) \ +{ \ + static gen_helper_zvdota_int * const fns[2] = { \ + gen_helper_##NAME##_b, NULL, \ + }; \ + if (CHECK(s, a)) { \ + TCGv_ptr dest, src1, src2, mask; \ + TCGv_i32 desc; \ + uint32_t data = 0; \ + \ + data = FIELD_DP32(data, VDATA, VM, a->vm); \ + data = FIELD_DP32(data, VDATA, LMUL, s->lmul); \ + data = FIELD_DP32(data, VDATA, VTA, s->vta); \ + data = FIELD_DP32(data, VDATA, VTA_ALL_1S, s->cfg_vta_all_1s); \ + \ + dest = tcg_temp_new_ptr(); \ + mask = tcg_temp_new_ptr(); \ + src1 = tcg_temp_new_ptr(); \ + src2 = tcg_temp_new_ptr(); \ + desc = tcg_constant_i32(simd_desc(s->cfg_ptr->vlenb, \ + s->cfg_ptr->vlenb, data)); \ + \ + tcg_gen_addi_ptr(dest, tcg_env, vreg_ofs(s, a->rd)); \ + tcg_gen_addi_ptr(src1, tcg_env, vreg_ofs(s, a->rs1)); \ + tcg_gen_addi_ptr(src2, tcg_env, vreg_ofs(s, a->rs2)); \ + tcg_gen_addi_ptr(mask, tcg_env, vreg_ofs(s, 0)); \ + \ + fns[s->sew](dest, mask, src1, src2, tcg_env, desc); \ + \ + finalize_rvv_inst(s); \ + \ + return true; \ + } \ + return false; \ +} + GEN_ABD_OPIVV_WIDEN_TRANS(vwabda_vv, vwabd_vv_check) GEN_ABD_OPIVX_WIDEN_TRANS(vwabda_vx, vwabd_vx_check) GEN_ABD_OPIVV_WIDEN_TRANS(vwabdau_vv, vwabd_vv_check) GEN_ABD_OPIVX_WIDEN_TRANS(vwabdau_vx, vwabd_vx_check) + +GEN_ZVDOTA_INT_TRANS(vqwdotau_vv, vext_vqwdota_check) +GEN_ZVDOTA_INT_TRANS(vqwdotas_vv, vext_vqwdota_check) diff --git a/target/riscv/tcg/vector_helper.c b/target/riscv/tcg/vector_helper.c index 4ba0a89f4b5..6abcf522648 100644 --- a/target/riscv/tcg/vector_helper.c +++ b/target/riscv/tcg/vector_helper.c @@ -5970,3 +5970,52 @@ GEN_VEXT_VV(vwabdau_vv_b, 2) GEN_VEXT_VV(vwabdau_vv_h, 4) GEN_VEXT_VX(vwabdau_vx_b, 2) GEN_VEXT_VX(vwabdau_vx_h, 4) + +/* + * Zvqwdota8i: 8-bit Integer Dot Product with Accumulation + */ +#define GEN_VEXT_ZVDOTA_INT(NAME, WD, W1, T2, HD, HS) \ +void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2, \ + CPURISCVState *env, uint32_t desc) \ +{ \ + uint32_t vm = vext_vm(desc); \ + uint32_t vl = env->vl; \ + uint32_t vta = vext_vta(desc); \ + uint32_t vta_all_1s = vext_vta_all_1s(desc); \ + uint16_t vlenb = riscv_cpu_cfg(env)->vlenb; \ + uint32_t total_elems = vlenb >> ctzl(sizeof(uint##WD##_t)); \ + uint8_t altfmt = FIELD_EX64(env->vtype, VTYPE, ALTFMT); \ + uint32_t i; \ + uint##WD##_t acc, dot_result, a; \ + \ + VSTART_CHECK_EARLY_EXIT(env, vl); \ + \ + acc = *((uint##WD##_t *)vd + HD(0)); \ + dot_result = 0; \ + \ + for (i = 0; i < vl; i++) { \ + if (!vm && !vext_elem_mask(v0, i)) { \ + continue; \ + } \ + \ + uint##WD##_t b = *((T2 *)vs2 + HS(i)); \ + if (altfmt) { \ + a = *((int##W1##_t *)vs1 + HS(i)); \ + } else { \ + a = *((uint##W1##_t *)vs1 + HS(i)); \ + } \ + dot_result += (a * b); \ + } \ + \ + *((uint##WD##_t *)vd + HD(0)) = acc + dot_result; \ + \ + env->vstart = 0; \ + \ + if (vta_all_1s) { \ + vext_set_elems_1s(vd, vta, sizeof(int##WD##_t), \ + total_elems * sizeof(int##WD##_t)); \ + } \ +} + +GEN_VEXT_ZVDOTA_INT(vqwdotau_vv_b, 32, 8, uint8_t, H4, H1) +GEN_VEXT_ZVDOTA_INT(vqwdotas_vv_b, 32, 8, int8_t, H4, H1) -- 2.56.0
