@@ -1422,3 +1422,25 @@ DEF_HELPER_6(th_vsxe_v_b, void, ptr, ptr, tl, ptr, env, i32)
DEF_HELPER_6(th_vsxe_v_h, void, ptr, ptr, tl, ptr, env, i32)
DEF_HELPER_6(th_vsxe_v_w, void, ptr, ptr, tl, ptr, env, i32)
DEF_HELPER_6(th_vsxe_v_d, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_5(th_vlbff_v_b, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlbff_v_h, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlbff_v_w, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlbff_v_d, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlhff_v_h, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlhff_v_w, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlhff_v_d, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlwff_v_w, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlwff_v_d, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vleff_v_b, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vleff_v_h, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vleff_v_w, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vleff_v_d, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlbuff_v_b, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlbuff_v_h, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlbuff_v_w, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlbuff_v_d, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlhuff_v_h, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlhuff_v_w, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlhuff_v_d, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlwuff_v_w, void, ptr, ptr, tl, env, i32)
+DEF_HELPER_5(th_vlwuff_v_d, void, ptr, ptr, tl, env, i32)
@@ -582,19 +582,62 @@ GEN_TH_TRANS(th_vsxh_v, 1, rnfvm, st_index_op_th, st_index_check_th)
GEN_TH_TRANS(th_vsxw_v, 2, rnfvm, st_index_op_th, st_index_check_th)
GEN_TH_TRANS(th_vsxe_v, 3, rnfvm, st_index_op_th, st_index_check_th)
+/*
+ * unit stride fault-only-first load
+ */
+
+/*
+ * This function is almost the copy of ldff_op, except:
+ * 1) different data encoding.
+ * 2) XTheadVector has more insns to handle zero/sign-extended.
+ */
+static bool ldff_op_th(DisasContext *s, arg_r2nfvm *a, uint8_t seq)
+{
+ uint32_t data = 0;
+ gen_helper_ldst_us_th *fn;
+ static gen_helper_ldst_us_th * const fns[7][4] = {
+ { gen_helper_th_vlbff_v_b, gen_helper_th_vlbff_v_h,
+ gen_helper_th_vlbff_v_w, gen_helper_th_vlbff_v_d },
+ { NULL, gen_helper_th_vlhff_v_h,
+ gen_helper_th_vlhff_v_w, gen_helper_th_vlhff_v_d },
+ { NULL, NULL,
+ gen_helper_th_vlwff_v_w, gen_helper_th_vlwff_v_d },
+ { gen_helper_th_vleff_v_b, gen_helper_th_vleff_v_h,
+ gen_helper_th_vleff_v_w, gen_helper_th_vleff_v_d },
+ { gen_helper_th_vlbuff_v_b, gen_helper_th_vlbuff_v_h,
+ gen_helper_th_vlbuff_v_w, gen_helper_th_vlbuff_v_d },
+ { NULL, gen_helper_th_vlhuff_v_h,
+ gen_helper_th_vlhuff_v_w, gen_helper_th_vlhuff_v_d },
+ { NULL, NULL,
+ gen_helper_th_vlwuff_v_w, gen_helper_th_vlwuff_v_d }
+ };
+
+ fn = fns[seq][s->sew];
+ if (fn == NULL) {
+ return false;
+ }
+
+ data = FIELD_DP32(data, VDATA_TH, MLEN, s->mlen);
+ data = FIELD_DP32(data, VDATA_TH, VM, a->vm);
+ data = FIELD_DP32(data, VDATA_TH, LMUL, s->lmul);
+ data = FIELD_DP32(data, VDATA_TH, NF, a->nf);
+ return ldff_trans(a->rd, a->rs1, data, fn, s);
+}
+
+GEN_TH_TRANS(th_vlbff_v, 0, r2nfvm, ldff_op_th, ld_us_check_th)
+GEN_TH_TRANS(th_vlhff_v, 1, r2nfvm, ldff_op_th, ld_us_check_th)
+GEN_TH_TRANS(th_vlwff_v, 2, r2nfvm, ldff_op_th, ld_us_check_th)
+GEN_TH_TRANS(th_vleff_v, 3, r2nfvm, ldff_op_th, ld_us_check_th)
+GEN_TH_TRANS(th_vlbuff_v, 4, r2nfvm, ldff_op_th, ld_us_check_th)
+GEN_TH_TRANS(th_vlhuff_v, 5, r2nfvm, ldff_op_th, ld_us_check_th)
+GEN_TH_TRANS(th_vlwuff_v, 6, r2nfvm, ldff_op_th, ld_us_check_th)
+
#define TH_TRANS_STUB(NAME) \
static bool trans_##NAME(DisasContext *s, arg_##NAME *a) \
{ \
return require_xtheadvector(s); \
}
-TH_TRANS_STUB(th_vlbff_v)
-TH_TRANS_STUB(th_vlhff_v)
-TH_TRANS_STUB(th_vlwff_v)
-TH_TRANS_STUB(th_vleff_v)
-TH_TRANS_STUB(th_vlbuff_v)
-TH_TRANS_STUB(th_vlhuff_v)
-TH_TRANS_STUB(th_vlwuff_v)
TH_TRANS_STUB(th_vamoswapw_v)
TH_TRANS_STUB(th_vamoaddw_v)
TH_TRANS_STUB(th_vamoxorw_v)
@@ -112,7 +112,7 @@ static inline uint32_t vext_max_elems(uint32_t desc, uint32_t log2_esz)
* and page table walk can't fill the TLB entry. Then the guest
* software can return here after process the exception or never return.
*/
-static void probe_pages(CPURISCVState *env, target_ulong addr,
+void probe_pages(CPURISCVState *env, target_ulong addr,
target_ulong len, uintptr_t ra,
MMUAccessType access_type)
{
@@ -243,4 +243,9 @@ target_ulong idx_b(target_ulong base, uint32_t idx, void *vs2);
target_ulong idx_h(target_ulong base, uint32_t idx, void *vs2);
target_ulong idx_w(target_ulong base, uint32_t idx, void *vs2);
target_ulong idx_d(target_ulong base, uint32_t idx, void *vs2);
+
+void probe_pages(CPURISCVState *env, target_ulong addr,
+ target_ulong len, uintptr_t ra,
+ MMUAccessType access_type);
+
#endif /* TARGET_RISCV_VECTOR_INTERNALS_H */
@@ -542,3 +542,122 @@ GEN_TH_ST_INDEX(th_vsxe_v_b, int8_t, int8_t, idx_b, ste_b)
GEN_TH_ST_INDEX(th_vsxe_v_h, int16_t, int16_t, idx_h, ste_h)
GEN_TH_ST_INDEX(th_vsxe_v_w, int32_t, int32_t, idx_w, ste_w)
GEN_TH_ST_INDEX(th_vsxe_v_d, int64_t, int64_t, idx_d, ste_d)
+
+/*
+ * unit-stride fault-only-first load instructions
+ */
+
+/*
+ * This function is almost the copy of vext_ldff, except:
+ * 1) different mask layout
+ * 2) different data encoding
+ * 3) different mask/tail elements process policy
+ */
+static inline void
+th_ldff(void *vd, void *v0, target_ulong base,
+ CPURISCVState *env, uint32_t desc,
+ th_ldst_elem_fn *ldst_elem,
+ clear_fn *clear_elem,
+ uint32_t esz, uint32_t msz, uintptr_t ra)
+{
+ void *host;
+ uint32_t i, k, vl = 0;
+ uint32_t mlen = th_mlen(desc);
+ uint32_t nf = th_nf(desc);
+ uint32_t vm = th_vm(desc);
+ uint32_t vlmax = th_maxsz(desc) / esz;
+ target_ulong addr, offset, remain;
+ int mmu_index = riscv_env_mmu_index(env, false);
+
+ VSTART_CHECK_EARLY_EXIT(env);
+ /* probe every access*/
+ for (i = env->vstart; i < env->vl; i++) {
+ if (!vm && !th_elem_mask(v0, mlen, i)) {
+ continue;
+ }
+ addr = adjust_addr(env, base + nf * i * msz);
+ if (i == 0) {
+ probe_pages(env, addr, nf * msz, ra, MMU_DATA_LOAD);
+ } else {
+ /* if it triggers an exception, no need to check watchpoint */
+ remain = nf * msz;
+ while (remain > 0) {
+ offset = -(addr | TARGET_PAGE_MASK);
+ host = tlb_vaddr_to_host(env, addr, MMU_DATA_LOAD, mmu_index);
+ if (host) {
+#ifdef CONFIG_USER_ONLY
+ if (!page_check_range(addr, offset, PAGE_READ)) {
+ vl = i;
+ goto ProbeSuccess;
+ }
+#else
+ probe_pages(env, addr, offset, ra, MMU_DATA_LOAD);
+#endif
+ } else {
+ vl = i;
+ goto ProbeSuccess;
+ }
+ if (remain <= offset) {
+ break;
+ }
+ remain -= offset;
+ addr = adjust_addr(env, addr + offset);
+ }
+ }
+ }
+ProbeSuccess:
+ /* load bytes from guest memory */
+ if (vl != 0) {
+ env->vl = vl;
+ }
+ for (i = env->vstart; i < env->vl; i++) {
+ k = 0;
+ if (!vm && !th_elem_mask(v0, mlen, i)) {
+ continue;
+ }
+ while (k < nf) {
+ addr = base + (i * nf + k) * msz;
+ ldst_elem(env, adjust_addr(env, addr), i + k * vlmax, vd, ra);
+ k++;
+ }
+ }
+ env->vstart = 0;
+ /* clear tail elements */
+ if (vl != 0) {
+ return;
+ }
+ for (k = 0; k < nf; k++) {
+ clear_elem(vd, env->vl + k * vlmax, env->vl * esz, vlmax * esz);
+ }
+}
+
+#define GEN_TH_LDFF(NAME, MTYPE, ETYPE, LOAD_FN, CLEAR_FN) \
+void HELPER(NAME)(void *vd, void *v0, target_ulong base, \
+ CPURISCVState *env, uint32_t desc) \
+{ \
+ th_ldff(vd, v0, base, env, desc, LOAD_FN, CLEAR_FN, \
+ sizeof(ETYPE), sizeof(MTYPE), GETPC()); \
+}
+
+GEN_TH_LDFF(th_vlbff_v_b, int8_t, int8_t, ldb_b, clearb_th)
+GEN_TH_LDFF(th_vlbff_v_h, int8_t, int16_t, ldb_h, clearh_th)
+GEN_TH_LDFF(th_vlbff_v_w, int8_t, int32_t, ldb_w, clearl_th)
+GEN_TH_LDFF(th_vlbff_v_d, int8_t, int64_t, ldb_d, clearq_th)
+GEN_TH_LDFF(th_vlhff_v_h, int16_t, int16_t, ldh_h, clearh_th)
+GEN_TH_LDFF(th_vlhff_v_w, int16_t, int32_t, ldh_w, clearl_th)
+GEN_TH_LDFF(th_vlhff_v_d, int16_t, int64_t, ldh_d, clearq_th)
+GEN_TH_LDFF(th_vlwff_v_w, int32_t, int32_t, ldw_w, clearl_th)
+GEN_TH_LDFF(th_vlwff_v_d, int32_t, int64_t, ldw_d, clearq_th)
+GEN_TH_LDFF(th_vleff_v_b, int8_t, int8_t, lde_b, clearb_th)
+GEN_TH_LDFF(th_vleff_v_h, int16_t, int16_t, lde_h, clearh_th)
+GEN_TH_LDFF(th_vleff_v_w, int32_t, int32_t, lde_w, clearl_th)
+GEN_TH_LDFF(th_vleff_v_d, int64_t, int64_t, lde_d, clearq_th)
+GEN_TH_LDFF(th_vlbuff_v_b, uint8_t, uint8_t, ldbu_b, clearb_th)
+GEN_TH_LDFF(th_vlbuff_v_h, uint8_t, uint16_t, ldbu_h, clearh_th)
+GEN_TH_LDFF(th_vlbuff_v_w, uint8_t, uint32_t, ldbu_w, clearl_th)
+GEN_TH_LDFF(th_vlbuff_v_d, uint8_t, uint64_t, ldbu_d, clearq_th)
+GEN_TH_LDFF(th_vlhuff_v_h, uint16_t, uint16_t, ldhu_h, clearh_th)
+GEN_TH_LDFF(th_vlhuff_v_w, uint16_t, uint32_t, ldhu_w, clearl_th)
+GEN_TH_LDFF(th_vlhuff_v_d, uint16_t, uint64_t, ldhu_d, clearq_th)
+GEN_TH_LDFF(th_vlwuff_v_w, uint32_t, uint32_t, ldwu_w, clearl_th)
+GEN_TH_LDFF(th_vlwuff_v_d, uint32_t, uint64_t, ldwu_d, clearq_th)
XTheadVector unit-stride fault-only-first instructions diff from RVV1.0 in the following points: 1. Different mask reg layout. 2. Different vector reg element width. 3. Different tail/masked elements process policy. 4. Different check policy. The detials of the difference are the same as unit-stride load instructions, as unit-stride fault-only-first instructions are the he special cases of unit-stride load operations. Signed-off-by: Huang Tao <eric.huang@linux.alibaba.com> --- target/riscv/helper.h | 22 ++++ .../riscv/insn_trans/trans_xtheadvector.c.inc | 57 +++++++-- target/riscv/vector_helper.c | 2 +- target/riscv/vector_internals.h | 5 + target/riscv/xtheadvector_helper.c | 119 ++++++++++++++++++ 5 files changed, 197 insertions(+), 8 deletions(-)