From 5ef5fdba17126067950d843830efea7c6b2eef12 Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Mon, 6 Mar 2023 18:58:15 -0800 Subject: Hexagon (target/hexagon) Add overrides for jumpr31 instructions Add overrides for SL2_jumpr31 Unconditional SL2_jumpr31_t Predicated true (old value) SL2_jumpr31_f Predicated false (old value) SL2_jumpr31_tnew Predicated true (new value) SL2_jumpr31_fnew Predicated false (new value) Signed-off-by: Taylor Simpson Reviewed-by: Anton Johansson Message-Id: <20230307025828.1612809-2-tsimpson@quicinc.com> --- target/hexagon/genptr.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) (limited to 'target/hexagon/genptr.c') diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index 86bd093..2bbe4e3 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -1,5 +1,5 @@ /* - * Copyright(c) 2019-2022 Qualcomm Innovation Center, Inc. All Rights Reserved. + * Copyright(c) 2019-2023 Qualcomm Innovation Center, Inc. All Rights Reserved. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -571,6 +571,13 @@ static void gen_cond_jumpr(DisasContext *ctx, TCGv dst_pc, gen_write_new_pc_addr(ctx, dst_pc, cond, pred); } +static void gen_cond_jumpr31(DisasContext *ctx, TCGCond cond, TCGv pred) +{ + TCGv LSB = tcg_temp_new(); + tcg_gen_andi_tl(LSB, pred, 1); + gen_cond_jumpr(ctx, hex_gpr[HEX_REG_LR], cond, LSB); +} + static void gen_cond_jump(DisasContext *ctx, TCGCond cond, TCGv pred, int pc_off) { -- cgit v1.1 From 242af2c0b3cc71d5b83fc81770b87d7c23808cd6 Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Mon, 6 Mar 2023 18:58:16 -0800 Subject: Hexagon (target/hexagon) Add overrides for callr Add overrides for J2_callr J2_callrt J2_callrf Signed-off-by: Taylor Simpson Reviewed-by: Anton Johansson Message-Id: <20230307025828.1612809-3-tsimpson@quicinc.com> --- target/hexagon/genptr.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) (limited to 'target/hexagon/genptr.c') diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index 2bbe4e3..592438f 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -682,6 +682,13 @@ static void gen_call(DisasContext *ctx, int pc_off) gen_write_new_pc_pcrel(ctx, pc_off, TCG_COND_ALWAYS, NULL); } +static void gen_callr(DisasContext *ctx, TCGv new_pc) +{ + TCGv next_PC = tcg_constant_tl(ctx->next_PC); + gen_log_reg_write(HEX_REG_LR, next_PC); + gen_write_new_pc_addr(ctx, new_pc, TCG_COND_ALWAYS, NULL); +} + static void gen_cond_call(DisasContext *ctx, TCGv pred, TCGCond cond, int pc_off) { @@ -697,6 +704,17 @@ static void gen_cond_call(DisasContext *ctx, TCGv pred, gen_set_label(skip); } +static void gen_cond_callr(DisasContext *ctx, + TCGCond cond, TCGv pred, TCGv new_pc) +{ + TCGv lsb = tcg_temp_new(); + TCGLabel *skip = gen_new_label(); + tcg_gen_andi_tl(lsb, pred, 1); + tcg_gen_brcondi_tl(cond, lsb, 0, skip); + gen_callr(ctx, new_pc); + gen_set_label(skip); +} + static void gen_endloop0(DisasContext *ctx) { TCGv lpcfg = tcg_temp_new(); -- cgit v1.1 From b8552a78a2e419b6680ade33fcf172f453b3d67b Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Mon, 6 Mar 2023 18:58:17 -0800 Subject: Hexagon (target/hexagon) Add overrides for endloop1/endloop01 Signed-off-by: Taylor Simpson Reviewed-by: Anton Johansson Message-Id: <20230307025828.1612809-4-tsimpson@quicinc.com> --- target/hexagon/genptr.c | 78 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 78 insertions(+) (limited to 'target/hexagon/genptr.c') diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index 592438f..f7017fd 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -770,6 +770,84 @@ static void gen_endloop0(DisasContext *ctx) } } +static void gen_endloop1(DisasContext *ctx) +{ + /* + * if (hex_gpr[HEX_REG_LC1] > 1) { + * PC = hex_gpr[HEX_REG_SA1]; + * hex_new_value[HEX_REG_LC1] = hex_gpr[HEX_REG_LC1] - 1; + * } + */ + TCGLabel *label = gen_new_label(); + tcg_gen_brcondi_tl(TCG_COND_LEU, hex_gpr[HEX_REG_LC1], 1, label); + { + gen_jumpr(ctx, hex_gpr[HEX_REG_SA1]); + tcg_gen_subi_tl(hex_new_value[HEX_REG_LC1], hex_gpr[HEX_REG_LC1], 1); + } + gen_set_label(label); +} + +static void gen_endloop01(DisasContext *ctx) +{ + TCGv lpcfg = tcg_temp_new(); + TCGLabel *label1 = gen_new_label(); + TCGLabel *label2 = gen_new_label(); + TCGLabel *label3 = gen_new_label(); + TCGLabel *done = gen_new_label(); + + GET_USR_FIELD(USR_LPCFG, lpcfg); + + /* + * if (lpcfg == 1) { + * hex_new_pred_value[3] = 0xff; + * hex_pred_written |= 1 << 3; + * } + */ + tcg_gen_brcondi_tl(TCG_COND_NE, lpcfg, 1, label1); + { + tcg_gen_movi_tl(hex_new_pred_value[3], 0xff); + tcg_gen_ori_tl(hex_pred_written, hex_pred_written, 1 << 3); + } + gen_set_label(label1); + + /* + * if (lpcfg) { + * SET_USR_FIELD(USR_LPCFG, lpcfg - 1); + * } + */ + tcg_gen_brcondi_tl(TCG_COND_EQ, lpcfg, 0, label2); + { + tcg_gen_subi_tl(lpcfg, lpcfg, 1); + SET_USR_FIELD(USR_LPCFG, lpcfg); + } + gen_set_label(label2); + + /* + * if (hex_gpr[HEX_REG_LC0] > 1) { + * PC = hex_gpr[HEX_REG_SA0]; + * hex_new_value[HEX_REG_LC0] = hex_gpr[HEX_REG_LC0] - 1; + * } else { + * if (hex_gpr[HEX_REG_LC1] > 1) { + * hex_next_pc = hex_gpr[HEX_REG_SA1]; + * hex_new_value[HEX_REG_LC1] = hex_gpr[HEX_REG_LC1] - 1; + * } + * } + */ + tcg_gen_brcondi_tl(TCG_COND_LEU, hex_gpr[HEX_REG_LC0], 1, label3); + { + gen_jumpr(ctx, hex_gpr[HEX_REG_SA0]); + tcg_gen_subi_tl(hex_new_value[HEX_REG_LC0], hex_gpr[HEX_REG_LC0], 1); + tcg_gen_br(done); + } + gen_set_label(label3); + tcg_gen_brcondi_tl(TCG_COND_LEU, hex_gpr[HEX_REG_LC1], 1, done); + { + gen_jumpr(ctx, hex_gpr[HEX_REG_SA1]); + tcg_gen_subi_tl(hex_new_value[HEX_REG_LC1], hex_gpr[HEX_REG_LC1], 1); + } + gen_set_label(done); +} + static void gen_cmp_jumpnv(DisasContext *ctx, TCGCond cond, TCGv val, TCGv src, int pc_off) { -- cgit v1.1 From dae386b80f27115fba6fd4f4ee215de8c6820e3b Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Mon, 6 Mar 2023 18:58:18 -0800 Subject: Hexagon (target/hexagon) Add overrides for dealloc-return instructions These instructions perform a deallocframe+return (jumpr r31) Add overrides for L4_return SL2_return L4_return_t L4_return_f L4_return_tnew_pt L4_return_fnew_pt L4_return_tnew_pnt L4_return_fnew_pnt SL2_return_t SL2_return_f SL2_return_tnew SL2_return_fnew This patch eliminates the last helper that uses write_new_pc, so we remove it from op_helper.c Signed-off-by: Taylor Simpson Reviewed-by: Anton Johansson Message-Id: <20230307025828.1612809-5-tsimpson@quicinc.com> --- target/hexagon/genptr.c | 77 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 77 insertions(+) (limited to 'target/hexagon/genptr.c') diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index f7017fd..3490310 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -715,6 +715,83 @@ static void gen_cond_callr(DisasContext *ctx, gen_set_label(skip); } +/* frame ^= (int64_t)FRAMEKEY << 32 */ +static void gen_frame_unscramble(TCGv_i64 frame) +{ + TCGv_i64 framekey = tcg_temp_new_i64(); + tcg_gen_extu_i32_i64(framekey, hex_gpr[HEX_REG_FRAMEKEY]); + tcg_gen_shli_i64(framekey, framekey, 32); + tcg_gen_xor_i64(frame, frame, framekey); +} + +static void gen_load_frame(DisasContext *ctx, TCGv_i64 frame, TCGv EA) +{ + Insn *insn = ctx->insn; /* Needed for CHECK_NOSHUF */ + CHECK_NOSHUF(EA, 8); + tcg_gen_qemu_ld64(frame, EA, ctx->mem_idx); +} + +static void gen_return_base(DisasContext *ctx, TCGv_i64 dst, TCGv src, + TCGv r29) +{ + /* + * frame = *src + * dst = frame_unscramble(frame) + * SP = src + 8 + * PC = dst.w[1] + */ + TCGv_i64 frame = tcg_temp_new_i64(); + TCGv r31 = tcg_temp_new(); + + gen_load_frame(ctx, frame, src); + gen_frame_unscramble(frame); + tcg_gen_mov_i64(dst, frame); + tcg_gen_addi_tl(r29, src, 8); + tcg_gen_extrh_i64_i32(r31, dst); + gen_jumpr(ctx, r31); +} + +static void gen_return(DisasContext *ctx, TCGv_i64 dst, TCGv src) +{ + TCGv r29 = tcg_temp_new(); + gen_return_base(ctx, dst, src, r29); + gen_log_reg_write(HEX_REG_SP, r29); +} + +/* if (pred) dst = dealloc_return(src):raw */ +static void gen_cond_return(DisasContext *ctx, TCGv_i64 dst, TCGv src, + TCGv pred, TCGCond cond) +{ + TCGv LSB = tcg_temp_new(); + TCGv mask = tcg_temp_new(); + TCGv r29 = tcg_temp_new(); + TCGLabel *skip = gen_new_label(); + tcg_gen_andi_tl(LSB, pred, 1); + + /* Initialize the results in case the predicate is false */ + tcg_gen_movi_i64(dst, 0); + tcg_gen_movi_tl(r29, 0); + + /* Set the bit in hex_slot_cancelled if the predicate is flase */ + tcg_gen_movi_tl(mask, 1 << ctx->insn->slot); + tcg_gen_or_tl(mask, hex_slot_cancelled, mask); + tcg_gen_movcond_tl(cond, hex_slot_cancelled, LSB, tcg_constant_tl(0), + mask, hex_slot_cancelled); + + tcg_gen_brcondi_tl(cond, LSB, 0, skip); + gen_return_base(ctx, dst, src, r29); + gen_set_label(skip); + gen_log_predicated_reg_write(HEX_REG_SP, r29, ctx->insn->slot); +} + +/* sub-instruction version (no RddV, so handle it manually) */ +static void gen_cond_return_subinsn(DisasContext *ctx, TCGCond cond, TCGv pred) +{ + TCGv_i64 RddV = tcg_temp_new_i64(); + gen_cond_return(ctx, RddV, hex_gpr[HEX_REG_FP], pred, cond); + gen_log_predicated_reg_write_pair(HEX_REG_FP, RddV, ctx->insn->slot); +} + static void gen_endloop0(DisasContext *ctx) { TCGv lpcfg = tcg_temp_new(); -- cgit v1.1 From 10849c2623af6f1c122956aaee8329b9414e637d Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Mon, 6 Mar 2023 18:58:19 -0800 Subject: Hexagon (target/hexagon) Analyze packet before generating TCG We create a new generator that creates an analyze_ function for each instruction. Currently, these functions record the writes to R, P, and C registers by calling ctx_log_reg_write[_pair] or ctx_log_pred_write. During gen_start_packet, we invoke the analyze_ function for each instruction in the packet, and we mark the implicit register and predicate writes. Doing the analysis up front has several advantages - We remove calls to ctx_log_* from gen_tcg_funcs.py and genptr.c - After the analysis is performed, we can initialize hex_new_value for each of the predicated assignments rather than during TCG generation for the instructions - This is a stepping stone for future work where the analysis will include the set of registers that are read. In cases where the packet doesn't have an overlap between the registers that are written and registers that are read, we can avoid the intermediate step of writing to hex_new_value. Note that other checks will also be needed (e.g., no instructions can raise an exception). Signed-off-by: Taylor Simpson Reviewed-by: Anton Johansson Message-Id: <20230307025828.1612809-6-tsimpson@quicinc.com> --- target/hexagon/genptr.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) (limited to 'target/hexagon/genptr.c') diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index 3490310..61ab9af 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -180,6 +180,7 @@ void gen_log_pred_write(DisasContext *ctx, int pnum, TCGv val) hex_new_pred_value[pnum], base_val); } tcg_gen_ori_tl(hex_pred_written, hex_pred_written, 1 << pnum); + set_bit(pnum, ctx->pregs_written); } static inline void gen_read_p3_0(TCGv control_reg) @@ -256,7 +257,6 @@ static void gen_write_p3_0(DisasContext *ctx, TCGv control_reg) for (int i = 0; i < NUM_PREGS; i++) { tcg_gen_extract_tl(hex_p8, control_reg, i * 8, 8); gen_log_pred_write(ctx, i, hex_p8); - ctx_log_pred_write(ctx, i); } } @@ -274,7 +274,6 @@ static inline void gen_write_ctrl_reg(DisasContext *ctx, int reg_num, gen_write_p3_0(ctx, val); } else { gen_log_reg_write(reg_num, val); - ctx_log_reg_write(ctx, reg_num); if (reg_num == HEX_REG_QEMU_PKT_CNT) { ctx->num_packets = 0; } @@ -296,10 +295,8 @@ static inline void gen_write_ctrl_reg_pair(DisasContext *ctx, int reg_num, gen_write_p3_0(ctx, val32); tcg_gen_extrh_i64_i32(val32, val); gen_log_reg_write(reg_num + 1, val32); - ctx_log_reg_write(ctx, reg_num + 1); } else { gen_log_reg_write_pair(reg_num, val); - ctx_log_reg_write_pair(ctx, reg_num); if (reg_num == HEX_REG_QEMU_PKT_CNT) { ctx->num_packets = 0; ctx->num_insns = 0; -- cgit v1.1 From e28b77a6b46bfec17ffb1f9764713b2c97418fb3 Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Mon, 6 Mar 2023 18:58:26 -0800 Subject: Hexagon (target/hexagon) Remove gen_log_predicated_reg_write[_pair] We assign the instruction destination register to hex_new_value[num] instead of a TCG temp that gets copied back to hex_new_value[num]. We introduce new functions get_result_gpr[_pair] to facilitate getting the proper destination register. Since we preload hex_new_value for predicated instructions, we don't need the check for slot_cancelled. So, we call gen_log_reg_write instead. We update the helper function generation and gen_tcg.h to maintain the disable-hexagon-idef-parser configuration. Here is a simple example of the differences in the TCG code generated: IN: 0x00400094: 0xf900c102 { if (P0) R2 = and(R0,R1) } BEFORE ---- 00400094 mov_i32 slot_cancelled,$0x0 mov_i32 new_r2,r2 mov_i32 loc2,$0x0 and_i32 tmp0,p0,$0x1 brcond_i32 tmp0,$0x0,eq,$L1 and_i32 tmp0,r0,r1 mov_i32 loc2,tmp0 br $L2 set_label $L1 or_i32 slot_cancelled,slot_cancelled,$0x8 set_label $L2 and_i32 tmp0,slot_cancelled,$0x8 movcond_i32 new_r2,tmp0,$0x0,loc2,new_r2,eq mov_i32 r2,new_r2 AFTER ---- 00400094 mov_i32 slot_cancelled,$0x0 mov_i32 new_r2,r2 and_i32 tmp0,p0,$0x1 brcond_i32 tmp0,$0x0,eq,$L1 and_i32 tmp0,r0,r1 mov_i32 new_r2,tmp0 br $L2 set_label $L1 or_i32 slot_cancelled,slot_cancelled,$0x8 set_label $L2 mov_i32 r2,new_r2 We'll remove the unnecessary manipulation of slot_cancelled in a subsequent patch. Signed-off-by: Taylor Simpson Reviewed-by: Anton Johansson Message-Id: <20230307025828.1612809-13-tsimpson@quicinc.com> --- target/hexagon/genptr.c | 121 ++++++++++++------------------------------------ 1 file changed, 30 insertions(+), 91 deletions(-) (limited to 'target/hexagon/genptr.c') diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index 61ab9af..da5dbc8 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -68,26 +68,17 @@ static inline void gen_masked_reg_write(TCGv new_val, TCGv cur_val, } } -static inline void gen_log_predicated_reg_write(int rnum, TCGv val, - uint32_t slot) +static TCGv get_result_gpr(DisasContext *ctx, int rnum) { - TCGv zero = tcg_constant_tl(0); - TCGv slot_mask = tcg_temp_new(); + return hex_new_value[rnum]; +} - tcg_gen_andi_tl(slot_mask, hex_slot_cancelled, 1 << slot); - tcg_gen_movcond_tl(TCG_COND_EQ, hex_new_value[rnum], slot_mask, zero, - val, hex_new_value[rnum]); - if (HEX_DEBUG) { - /* - * Do this so HELPER(debug_commit_end) will know - * - * Note that slot_mask indicates the value is not written - * (i.e., slot was cancelled), so we create a true/false value before - * or'ing with hex_reg_written[rnum]. - */ - tcg_gen_setcond_tl(TCG_COND_EQ, slot_mask, slot_mask, zero); - tcg_gen_or_tl(hex_reg_written[rnum], hex_reg_written[rnum], slot_mask); - } +static TCGv_i64 get_result_gpr_pair(DisasContext *ctx, int rnum) +{ + TCGv_i64 result = tcg_temp_new_i64(); + tcg_gen_concat_i32_i64(result, hex_new_value[rnum], + hex_new_value[rnum + 1]); + return result; } void gen_log_reg_write(int rnum, TCGv val) @@ -102,39 +93,6 @@ void gen_log_reg_write(int rnum, TCGv val) } } -static void gen_log_predicated_reg_write_pair(int rnum, TCGv_i64 val, - uint32_t slot) -{ - TCGv val32 = tcg_temp_new(); - TCGv zero = tcg_constant_tl(0); - TCGv slot_mask = tcg_temp_new(); - - tcg_gen_andi_tl(slot_mask, hex_slot_cancelled, 1 << slot); - /* Low word */ - tcg_gen_extrl_i64_i32(val32, val); - tcg_gen_movcond_tl(TCG_COND_EQ, hex_new_value[rnum], - slot_mask, zero, - val32, hex_new_value[rnum]); - /* High word */ - tcg_gen_extrh_i64_i32(val32, val); - tcg_gen_movcond_tl(TCG_COND_EQ, hex_new_value[rnum + 1], - slot_mask, zero, - val32, hex_new_value[rnum + 1]); - if (HEX_DEBUG) { - /* - * Do this so HELPER(debug_commit_end) will know - * - * Note that slot_mask indicates the value is not written - * (i.e., slot was cancelled), so we create a true/false value before - * or'ing with hex_reg_written[rnum]. - */ - tcg_gen_setcond_tl(TCG_COND_EQ, slot_mask, slot_mask, zero); - tcg_gen_or_tl(hex_reg_written[rnum], hex_reg_written[rnum], slot_mask); - tcg_gen_or_tl(hex_reg_written[rnum + 1], hex_reg_written[rnum + 1], - slot_mask); - } -} - static void gen_log_reg_write_pair(int rnum, TCGv_i64 val) { const target_ulong reg_mask_low = reg_immut_masks[rnum]; @@ -290,11 +248,12 @@ static inline void gen_write_ctrl_reg_pair(DisasContext *ctx, int reg_num, TCGv_i64 val) { if (reg_num == HEX_REG_P3_0_ALIASED) { + TCGv result = get_result_gpr(ctx, reg_num + 1); TCGv val32 = tcg_temp_new(); tcg_gen_extrl_i64_i32(val32, val); gen_write_p3_0(ctx, val32); tcg_gen_extrh_i64_i32(val32, val); - gen_log_reg_write(reg_num + 1, val32); + tcg_gen_mov_tl(result, val32); } else { gen_log_reg_write_pair(reg_num, val); if (reg_num == HEX_REG_QEMU_PKT_CNT) { @@ -673,31 +632,28 @@ static void gen_jumpr(DisasContext *ctx, TCGv new_pc) static void gen_call(DisasContext *ctx, int pc_off) { - TCGv next_PC = - tcg_constant_tl(ctx->pkt->pc + ctx->pkt->encod_pkt_size_in_bytes); - gen_log_reg_write(HEX_REG_LR, next_PC); + TCGv lr = get_result_gpr(ctx, HEX_REG_LR); + tcg_gen_movi_tl(lr, ctx->next_PC); gen_write_new_pc_pcrel(ctx, pc_off, TCG_COND_ALWAYS, NULL); } static void gen_callr(DisasContext *ctx, TCGv new_pc) { - TCGv next_PC = tcg_constant_tl(ctx->next_PC); - gen_log_reg_write(HEX_REG_LR, next_PC); + TCGv lr = get_result_gpr(ctx, HEX_REG_LR); + tcg_gen_movi_tl(lr, ctx->next_PC); gen_write_new_pc_addr(ctx, new_pc, TCG_COND_ALWAYS, NULL); } static void gen_cond_call(DisasContext *ctx, TCGv pred, TCGCond cond, int pc_off) { - TCGv next_PC; + TCGv lr = get_result_gpr(ctx, HEX_REG_LR); TCGv lsb = tcg_temp_new(); TCGLabel *skip = gen_new_label(); tcg_gen_andi_tl(lsb, pred, 1); gen_write_new_pc_pcrel(ctx, pc_off, cond, lsb); tcg_gen_brcondi_tl(cond, lsb, 0, skip); - next_PC = - tcg_constant_tl(ctx->pkt->pc + ctx->pkt->encod_pkt_size_in_bytes); - gen_log_reg_write(HEX_REG_LR, next_PC); + tcg_gen_movi_tl(lr, ctx->next_PC); gen_set_label(skip); } @@ -728,8 +684,7 @@ static void gen_load_frame(DisasContext *ctx, TCGv_i64 frame, TCGv EA) tcg_gen_qemu_ld64(frame, EA, ctx->mem_idx); } -static void gen_return_base(DisasContext *ctx, TCGv_i64 dst, TCGv src, - TCGv r29) +static void gen_return(DisasContext *ctx, TCGv_i64 dst, TCGv src) { /* * frame = *src @@ -739,6 +694,7 @@ static void gen_return_base(DisasContext *ctx, TCGv_i64 dst, TCGv src, */ TCGv_i64 frame = tcg_temp_new_i64(); TCGv r31 = tcg_temp_new(); + TCGv r29 = get_result_gpr(ctx, HEX_REG_SP); gen_load_frame(ctx, frame, src); gen_frame_unscramble(frame); @@ -748,45 +704,25 @@ static void gen_return_base(DisasContext *ctx, TCGv_i64 dst, TCGv src, gen_jumpr(ctx, r31); } -static void gen_return(DisasContext *ctx, TCGv_i64 dst, TCGv src) -{ - TCGv r29 = tcg_temp_new(); - gen_return_base(ctx, dst, src, r29); - gen_log_reg_write(HEX_REG_SP, r29); -} - /* if (pred) dst = dealloc_return(src):raw */ static void gen_cond_return(DisasContext *ctx, TCGv_i64 dst, TCGv src, TCGv pred, TCGCond cond) { TCGv LSB = tcg_temp_new(); - TCGv mask = tcg_temp_new(); - TCGv r29 = tcg_temp_new(); TCGLabel *skip = gen_new_label(); tcg_gen_andi_tl(LSB, pred, 1); - /* Initialize the results in case the predicate is false */ - tcg_gen_movi_i64(dst, 0); - tcg_gen_movi_tl(r29, 0); - - /* Set the bit in hex_slot_cancelled if the predicate is flase */ - tcg_gen_movi_tl(mask, 1 << ctx->insn->slot); - tcg_gen_or_tl(mask, hex_slot_cancelled, mask); - tcg_gen_movcond_tl(cond, hex_slot_cancelled, LSB, tcg_constant_tl(0), - mask, hex_slot_cancelled); - tcg_gen_brcondi_tl(cond, LSB, 0, skip); - gen_return_base(ctx, dst, src, r29); + gen_return(ctx, dst, src); gen_set_label(skip); - gen_log_predicated_reg_write(HEX_REG_SP, r29, ctx->insn->slot); } /* sub-instruction version (no RddV, so handle it manually) */ static void gen_cond_return_subinsn(DisasContext *ctx, TCGCond cond, TCGv pred) { - TCGv_i64 RddV = tcg_temp_new_i64(); + TCGv_i64 RddV = get_result_gpr_pair(ctx, HEX_REG_FP); gen_cond_return(ctx, RddV, hex_gpr[HEX_REG_FP], pred, cond); - gen_log_predicated_reg_write_pair(HEX_REG_FP, RddV, ctx->insn->slot); + gen_log_reg_write_pair(HEX_REG_FP, RddV); } static void gen_endloop0(DisasContext *ctx) @@ -836,9 +772,9 @@ static void gen_endloop0(DisasContext *ctx) TCGLabel *label3 = gen_new_label(); tcg_gen_brcondi_tl(TCG_COND_LEU, hex_gpr[HEX_REG_LC0], 1, label3); { + TCGv lc0 = get_result_gpr(ctx, HEX_REG_LC0); gen_jumpr(ctx, hex_gpr[HEX_REG_SA0]); - tcg_gen_subi_tl(hex_new_value[HEX_REG_LC0], - hex_gpr[HEX_REG_LC0], 1); + tcg_gen_subi_tl(lc0, hex_gpr[HEX_REG_LC0], 1); } gen_set_label(label3); } @@ -855,8 +791,9 @@ static void gen_endloop1(DisasContext *ctx) TCGLabel *label = gen_new_label(); tcg_gen_brcondi_tl(TCG_COND_LEU, hex_gpr[HEX_REG_LC1], 1, label); { + TCGv lc1 = get_result_gpr(ctx, HEX_REG_LC1); gen_jumpr(ctx, hex_gpr[HEX_REG_SA1]); - tcg_gen_subi_tl(hex_new_value[HEX_REG_LC1], hex_gpr[HEX_REG_LC1], 1); + tcg_gen_subi_tl(lc1, hex_gpr[HEX_REG_LC1], 1); } gen_set_label(label); } @@ -909,15 +846,17 @@ static void gen_endloop01(DisasContext *ctx) */ tcg_gen_brcondi_tl(TCG_COND_LEU, hex_gpr[HEX_REG_LC0], 1, label3); { + TCGv lc0 = get_result_gpr(ctx, HEX_REG_LC0); gen_jumpr(ctx, hex_gpr[HEX_REG_SA0]); - tcg_gen_subi_tl(hex_new_value[HEX_REG_LC0], hex_gpr[HEX_REG_LC0], 1); + tcg_gen_subi_tl(lc0, hex_gpr[HEX_REG_LC0], 1); tcg_gen_br(done); } gen_set_label(label3); tcg_gen_brcondi_tl(TCG_COND_LEU, hex_gpr[HEX_REG_LC1], 1, done); { + TCGv lc1 = get_result_gpr(ctx, HEX_REG_LC1); gen_jumpr(ctx, hex_gpr[HEX_REG_SA1]); - tcg_gen_subi_tl(hex_new_value[HEX_REG_LC1], hex_gpr[HEX_REG_LC1], 1); + tcg_gen_subi_tl(lc1, hex_gpr[HEX_REG_LC1], 1); } gen_set_label(done); } -- cgit v1.1 From c2b33d0be998bf539953f1dad0aa0d1cc8d9d069 Mon Sep 17 00:00:00 2001 From: Taylor Simpson Date: Mon, 6 Mar 2023 18:58:28 -0800 Subject: Hexagon (target/hexagon) Improve code gen for predicated HVX instructions The following improvements are made for predicated HVX instructions During gen_commit_hvx, unconditionally move the "new" value into the dest Don't set slot_cancelled Remove runtime bookkeeping of which registers were updated Reduce the cases where gen_log_vreg_write[_pair] is called It's only needed for special operands VxxV and VyV Remove gen_log_qreg_write Signed-off-by: Taylor Simpson Reviewed-by: Anton Johansson Message-Id: <20230307025828.1612809-15-tsimpson@quicinc.com> --- target/hexagon/genptr.c | 48 ++++++------------------------------------------ 1 file changed, 6 insertions(+), 42 deletions(-) (limited to 'target/hexagon/genptr.c') diff --git a/target/hexagon/genptr.c b/target/hexagon/genptr.c index da5dbc8..bb274d4 100644 --- a/target/hexagon/genptr.c +++ b/target/hexagon/genptr.c @@ -985,68 +985,32 @@ static intptr_t vreg_src_off(DisasContext *ctx, int num) } static void gen_log_vreg_write(DisasContext *ctx, intptr_t srcoff, int num, - VRegWriteType type, int slot_num, - bool is_predicated) + VRegWriteType type) { - TCGLabel *label_end = NULL; intptr_t dstoff; - if (is_predicated) { - TCGv cancelled = tcg_temp_new(); - label_end = gen_new_label(); - - /* Don't do anything if the slot was cancelled */ - tcg_gen_extract_tl(cancelled, hex_slot_cancelled, slot_num, 1); - tcg_gen_brcondi_tl(TCG_COND_NE, cancelled, 0, label_end); - } - if (type != EXT_TMP) { dstoff = ctx_future_vreg_off(ctx, num, 1, true); tcg_gen_gvec_mov(MO_64, dstoff, srcoff, sizeof(MMVector), sizeof(MMVector)); - tcg_gen_ori_tl(hex_VRegs_updated, hex_VRegs_updated, 1 << num); } else { dstoff = ctx_tmp_vreg_off(ctx, num, 1, false); tcg_gen_gvec_mov(MO_64, dstoff, srcoff, sizeof(MMVector), sizeof(MMVector)); } - - if (is_predicated) { - gen_set_label(label_end); - } } static void gen_log_vreg_write_pair(DisasContext *ctx, intptr_t srcoff, int num, - VRegWriteType type, int slot_num, - bool is_predicated) + VRegWriteType type) { - gen_log_vreg_write(ctx, srcoff, num ^ 0, type, slot_num, is_predicated); + gen_log_vreg_write(ctx, srcoff, num ^ 0, type); srcoff += sizeof(MMVector); - gen_log_vreg_write(ctx, srcoff, num ^ 1, type, slot_num, is_predicated); + gen_log_vreg_write(ctx, srcoff, num ^ 1, type); } -static void gen_log_qreg_write(intptr_t srcoff, int num, int vnew, - int slot_num, bool is_predicated) +static intptr_t get_result_qreg(DisasContext *ctx, int qnum) { - TCGLabel *label_end = NULL; - intptr_t dstoff; - - if (is_predicated) { - TCGv cancelled = tcg_temp_new(); - label_end = gen_new_label(); - - /* Don't do anything if the slot was cancelled */ - tcg_gen_extract_tl(cancelled, hex_slot_cancelled, slot_num, 1); - tcg_gen_brcondi_tl(TCG_COND_NE, cancelled, 0, label_end); - } - - dstoff = offsetof(CPUHexagonState, future_QRegs[num]); - tcg_gen_gvec_mov(MO_64, dstoff, srcoff, sizeof(MMQReg), sizeof(MMQReg)); - - if (is_predicated) { - tcg_gen_ori_tl(hex_QRegs_updated, hex_QRegs_updated, 1 << num); - gen_set_label(label_end); - } + return offsetof(CPUHexagonState, future_QRegs[qnum]); } static void gen_vreg_load(DisasContext *ctx, intptr_t dstoff, TCGv src, -- cgit v1.1