Add hex_next_PC, a global mirroring CPUHexagonState::next_PC, and ctx->need_next_pc, so that gen_write_new_pc_addr() can write the branch target through hex_next_PC instead of hex_gpr[HEX_REG_PC] when a later unconditional write of PC is expected. gen_end_tb() then commits hex_next_PC into hex_gpr[HEX_REG_PC] at the end of the packet.
Previously, non-COF instructions that still end a TB did not advance the PC, since next_PC's value was never written back into the PC register. Signed-off-by: Brian Cain <[email protected]> --- target/hexagon/translate.h | 3 +++ target/hexagon/genptr.c | 14 ++++++++--- target/hexagon/translate.c | 51 ++++++++++++++++++++++++++------------ 3 files changed, 49 insertions(+), 19 deletions(-) diff --git a/target/hexagon/translate.h b/target/hexagon/translate.h index 0a7f37d5842..2fca157553b 100644 --- a/target/hexagon/translate.h +++ b/target/hexagon/translate.h @@ -40,6 +40,7 @@ typedef struct DisasContext { int reg_log_idx; DECLARE_BITMAP(regs_written, TOTAL_PER_THREAD_REGS); DECLARE_BITMAP(predicated_regs, TOTAL_PER_THREAD_REGS); + bool pkt_ends_tb; bool implicit_usr_write; #ifndef CONFIG_USER_ONLY int greg_log[GREG_WRITES_MAX]; @@ -75,6 +76,7 @@ typedef struct DisasContext { DECLARE_BITMAP(insn_qregs_read, NUM_QREGS); bool pre_commit; bool need_commit; + bool need_next_pc; TCGCond branch_cond; target_ulong branch_dest; bool is_tight_loop; @@ -310,6 +312,7 @@ extern TCGv hex_gpr[TOTAL_PER_THREAD_REGS]; extern TCGv hex_pred[NUM_PREGS]; extern TCGv hex_slot_cancelled; extern TCGv hex_new_value_usr; +extern TCGv hex_next_PC; extern TCGv hex_store_addr[STORES_MAX]; extern TCGv_i32 hex_store_width[STORES_MAX]; extern TCGv hex_store_val32[STORES_MAX]; diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index 1ebc7471409..dc9ba8a0060 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -613,14 +613,22 @@ static void gen_write_new_pc_addr(DisasContext *ctx, TCGv addr, tcg_gen_brcondi_tl(cond, pred, 1, pred_false); } + /* + * If gen_end_tb() will unconditionally overwrite PC with hex_next_PC + * (because this packet has a predicated COF that may not execute), + * write the branch target there instead of directly into the PC + * global, or the overwrite in gen_end_tb() would clobber it. + */ + TCGv pc_wr = ctx->need_next_pc ? hex_next_PC : hex_gpr[HEX_REG_PC]; + if (ctx->pkt.pkt_has_multi_cof) { /* If there are multiple branches in a packet, ignore the second one */ - tcg_gen_movcond_tl(TCG_COND_NE, hex_gpr[HEX_REG_PC], + tcg_gen_movcond_tl(TCG_COND_NE, pc_wr, ctx->branch_taken, tcg_constant_tl(0), - hex_gpr[HEX_REG_PC], addr); + pc_wr, addr); tcg_gen_movi_tl(ctx->branch_taken, 1); } else { - tcg_gen_mov_tl(hex_gpr[HEX_REG_PC], addr); + tcg_gen_mov_tl(pc_wr, addr); } if (cond != TCG_COND_ALWAYS) { diff --git a/target/hexagon/translate.c b/target/hexagon/translate.c index 28a171bf623..22f0ff5b891 100644 --- a/target/hexagon/translate.c +++ b/target/hexagon/translate.c @@ -54,6 +54,7 @@ static const AnalyzeInsn opcode_analyze[XX_LAST_OPCODE] = { TCGv hex_gpr[TOTAL_PER_THREAD_REGS]; TCGv hex_pred[NUM_PREGS]; TCGv hex_slot_cancelled; +TCGv hex_next_PC; TCGv hex_new_value_usr; TCGv hex_store_addr[STORES_MAX]; TCGv_i32 hex_store_width[STORES_MAX]; @@ -184,10 +185,16 @@ static void gen_goto_tb(DisasContext *ctx, unsigned tb_slot_idx, } } +static bool need_next_PC(DisasContext *ctx); + static void gen_end_tb(DisasContext *ctx) { gen_exec_counters(ctx); + if (ctx->need_next_pc) { + tcg_gen_mov_tl(hex_gpr[HEX_REG_PC], hex_next_PC); + } + if (ctx->branch_cond != TCG_COND_NEVER) { if (ctx->branch_cond != TCG_COND_ALWAYS) { TCGLabel *skip = gen_new_label(); @@ -391,17 +398,25 @@ static bool pkt_ends_tb(Packet *pkt) static bool need_next_PC(DisasContext *ctx) { - /* Check for conditional control flow or HW loop end */ - for (int i = 0; i < ctx->pkt.num_insns; i++) { - uint16_t opcode = ctx->pkt.insn[i].opcode; - if (GET_ATTRIB(opcode, A_CONDEXEC) && GET_ATTRIB(opcode, A_COF)) { - return true; - } - if (GET_ATTRIB(opcode, A_HWLOOP0_END) || - GET_ATTRIB(opcode, A_HWLOOP1_END)) { - return true; + Packet *pkt = &ctx->pkt; + if (pkt->pkt_has_cof || ctx->pkt_ends_tb) { + for (int i = 0; i < pkt->num_insns; i++) { + uint16_t opcode = pkt->insn[i].opcode; + if ((GET_ATTRIB(opcode, A_CONDEXEC) && GET_ATTRIB(opcode, A_COF)) || + GET_ATTRIB(opcode, A_HWLOOP0_END) || + GET_ATTRIB(opcode, A_HWLOOP1_END)) { + return true; + } } } + /* + * We end the TB on some instructions that do not change the flow (for + * other reasons). In these cases, we must set pc too, as the insn won't + * do it themselves. + */ + if (ctx->pkt_ends_tb && !check_for_attrib(pkt, A_COF)) { + return true; + } return false; } @@ -637,12 +652,14 @@ static void gen_start_packet(DisasContext *ctx) ctx->branch_taken = NULL; if (ctx->pkt.pkt_has_cof) { ctx->branch_taken = tcg_temp_new(); - if (ctx->pkt.pkt_has_multi_cof) { - tcg_gen_movi_tl(ctx->branch_taken, 0); - } - if (need_next_PC(ctx)) { - tcg_gen_movi_tl(hex_gpr[HEX_REG_PC], next_PC); - } + } + if (ctx->pkt.pkt_has_multi_cof) { + tcg_gen_movi_tl(ctx->branch_taken, 0); + } + ctx->pkt_ends_tb = pkt_ends_tb(&ctx->pkt); + ctx->need_next_pc = need_next_PC(ctx); + if (ctx->need_next_pc) { + tcg_gen_movi_tl(hex_next_PC, next_PC); } /* Preload the predicated registers into get_result_gpr(ctx, i) */ @@ -1142,7 +1159,7 @@ static void gen_commit_packet(DisasContext *ctx) ctx->pkt.vhist_insn->generate(ctx); } - if (pkt_ends_tb(&ctx->pkt) || ctx->base.is_jmp == DISAS_NORETURN) { + if (ctx->pkt_ends_tb || ctx->base.is_jmp == DISAS_NORETURN) { gen_end_tb(ctx); } } @@ -1327,6 +1344,8 @@ void hexagon_translate_init(void) } hex_new_value_usr = tcg_global_mem_new(tcg_env, offsetof(CPUHexagonState, new_value_usr), "new_value_usr"); + hex_next_PC = tcg_global_mem_new(tcg_env, + offsetof(CPUHexagonState, next_PC), "next_PC"); for (i = 0; i < NUM_PREGS; i++) { hex_pred[i] = tcg_global_mem_new(tcg_env, -- 2.34.1
