From: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
To: Huang Tao <eric.huang@linux.alibaba.com>, qemu-devel@nongnu.org
Cc: qemu-riscv@nongnu.org, zhiwei_liu@linux.alibaba.com,
liwei1518@gmail.com, bin.meng@windriver.com,
alistair.francis@wdc.com, palmer@dabbelt.com,
Christoph Muellner <christoph.muellner@vrull.eu>,
Richard Henderson <richard.henderson@linaro.org>
Subject: Re: [PATCH] target/riscv: Implement dynamic establishment of custom decoder
Date: Thu, 7 Mar 2024 16:55:25 -0300 [thread overview]
Message-ID: <9b9c1af9-5dca-4270-8dfe-a62223c8cbbb@ventanamicro.com> (raw)
In-Reply-To: <20240306093308.35188-1-eric.huang@linux.alibaba.com>
(--- adding Richard ---)
On 3/6/24 06:33, Huang Tao wrote:
> In this patch, we modify the decoder to be a freely composable data
> structure instead of a hardcoded one. It can be dynamically builded up
> according to the extensions.
> This approach has several benefits:
> 1. Provides support for heterogeneous cpu architectures. As we add decoder in
> CPUArchState, each cpu can have their own decoder, and the decoders can be
> different due to cpu's features.
> 2. Improve the decoding efficiency. We run the guard_func to see if the decoder
> can be added to the dynamic_decoder when building up the decoder. Therefore,
> there is no need to run the guard_func when decoding each instruction. It can
> improve the decoding efficiency
> 3. For vendor or dynamic cpus, it allows them to customize their own decoder
> functions to improve decoding efficiency, especially when vendor-defined
> instruction sets increase. Because of dynamic building up, it can skip the other
> decoder guard functions when decoding.
>
> Signed-off-by: Huang Tao <eric.huang@linux.alibaba.com>
> Suggested-by: Christoph Muellner <christoph.muellner@vrull.eu>
> Co-authored-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
> ---
> target/riscv/cpu.c | 20 ++++++++++++++++++++
> target/riscv/cpu.h | 2 ++
> target/riscv/cpu_decoder.h | 34 ++++++++++++++++++++++++++++++++++
> target/riscv/translate.c | 28 ++++++++++++----------------
> 4 files changed, 68 insertions(+), 16 deletions(-)
> create mode 100644 target/riscv/cpu_decoder.h
>
> diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
> index 5ff0192c52..cadcd51b4f 100644
> --- a/target/riscv/cpu.c
> +++ b/target/riscv/cpu.c
> @@ -38,6 +38,7 @@
> #include "kvm/kvm_riscv.h"
> #include "tcg/tcg-cpu.h"
> #include "tcg/tcg.h"
> +#include "cpu_decoder.h"
>
> /* RISC-V CPU definitions */
> static const char riscv_single_letter_exts[] = "IEMAFDQCBPVH";
> @@ -1102,6 +1103,23 @@ static void riscv_cpu_satp_mode_finalize(RISCVCPU *cpu, Error **errp)
> }
> #endif
>
> +static void riscv_cpu_finalize_dynamic_decoder(RISCVCPU *cpu)
> +{
> + CPURISCVState *env = &cpu->env;
> + decode_fn *dynamic_decoder;
> + dynamic_decoder = g_new0(decode_fn, decoder_table_size);
> + int j = 0;
> + for (size_t i = 0; i < decoder_table_size; ++i) {
> + if (decoder_table[i].guard_func &&
> + decoder_table[i].guard_func(&cpu->cfg)) {
> + dynamic_decoder[j] = decoder_table[i].decode_fn;
> + j++;
> + }
> + }
> +
> + env->decoder = dynamic_decoder;
> +}
> +
> void riscv_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
> {
> Error *local_err = NULL;
> @@ -1127,6 +1145,8 @@ void riscv_cpu_finalize_features(RISCVCPU *cpu, Error **errp)
> return;
> }
> }
> +
> + riscv_cpu_finalize_dynamic_decoder(cpu);
> }
>
> static void riscv_cpu_realize(DeviceState *dev, Error **errp)
> diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
> index 5d291a7092..42bfed065c 100644
> --- a/target/riscv/cpu.h
> +++ b/target/riscv/cpu.h
> @@ -30,6 +30,7 @@
> #include "qemu/int128.h"
> #include "cpu_bits.h"
> #include "cpu_cfg.h"
> +#include "cpu_decoder.h"
> #include "qapi/qapi-types-common.h"
> #include "cpu-qom.h"
>
> @@ -433,6 +434,7 @@ struct CPUArchState {
> uint64_t kvm_timer_state;
> uint64_t kvm_timer_frequency;
> #endif /* CONFIG_KVM */
> + const decode_fn *decoder;
> };
>
> /*
> diff --git a/target/riscv/cpu_decoder.h b/target/riscv/cpu_decoder.h
> new file mode 100644
> index 0000000000..549414ce4c
> --- /dev/null
> +++ b/target/riscv/cpu_decoder.h
> @@ -0,0 +1,34 @@
> +/*
> + * QEMU RISC-V CPU Decoder
> + *
> + * Copyright (c) 2023-2024 Alibaba Group
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms and conditions of the GNU General Public License,
> + * version 2 or later, as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope it will be useful, but WITHOUT
> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
> + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
> + * more details.
> + *
> + * You should have received a copy of the GNU General Public License along with
> + * this program. If not, see <http://www.gnu.org/licenses/>.
> + */
> +
> +#ifndef RISCV_CPU_DECODER_H
> +#define RISCV_CPU_DECODER_H
> +
> +struct DisasContext;
> +struct RISCVCPUConfig;
> +typedef struct RISCVDecoder {
> + bool (*guard_func)(const struct RISCVCPUConfig *);
> + bool (*decode_fn)(struct DisasContext *, uint32_t);
> +} RISCVDecoder;
> +
> +typedef bool (*decode_fn)(struct DisasContext *, uint32_t);
> +
> +extern const size_t decoder_table_size;
> +
> +extern const RISCVDecoder decoder_table[];
> +#endif
> diff --git a/target/riscv/translate.c b/target/riscv/translate.c
> index 177418b2b9..23c5321bce 100644
> --- a/target/riscv/translate.c
> +++ b/target/riscv/translate.c
> @@ -115,6 +115,7 @@ typedef struct DisasContext {
> bool frm_valid;
> /* TCG of the current insn_start */
> TCGOp *insn_start;
> + const decode_fn *decoder;
> } DisasContext;
>
> static inline bool has_ext(DisasContext *ctx, uint32_t ext)
> @@ -1118,21 +1119,16 @@ static inline int insn_len(uint16_t first_word)
> return (first_word & 3) == 3 ? 4 : 2;
> }
>
> +const RISCVDecoder decoder_table[] = {
> + { always_true_p, decode_insn32 },
> + { has_xthead_p, decode_xthead},
> + { has_XVentanaCondOps_p, decode_XVentanaCodeOps},
> +};
> +
> +const size_t decoder_table_size = ARRAY_SIZE(decoder_table);
> +
> static void decode_opc(CPURISCVState *env, DisasContext *ctx, uint16_t opcode)
> {
> - /*
> - * A table with predicate (i.e., guard) functions and decoder functions
> - * that are tested in-order until a decoder matches onto the opcode.
> - */
> - static const struct {
> - bool (*guard_func)(const RISCVCPUConfig *);
> - bool (*decode_func)(DisasContext *, uint32_t);
> - } decoders[] = {
> - { always_true_p, decode_insn32 },
> - { has_xthead_p, decode_xthead },
> - { has_XVentanaCondOps_p, decode_XVentanaCodeOps },
> - };
> -
> ctx->virt_inst_excp = false;
> ctx->cur_insn_len = insn_len(opcode);
> /* Check for compressed insn */
> @@ -1153,9 +1149,8 @@ static void decode_opc(CPURISCVState *env, DisasContext *ctx, uint16_t opcode)
> ctx->base.pc_next + 2));
> ctx->opcode = opcode32;
>
> - for (size_t i = 0; i < ARRAY_SIZE(decoders); ++i) {
> - if (decoders[i].guard_func(ctx->cfg_ptr) &&
> - decoders[i].decode_func(ctx, opcode32)) {
> + for (size_t i = 0; i < decoder_table_size; ++i) {
> + if (ctx->decoder[i](ctx, opcode32)) {
> return;
> }
> }
> @@ -1199,6 +1194,7 @@ static void riscv_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
> ctx->itrigger = FIELD_EX32(tb_flags, TB_FLAGS, ITRIGGER);
> ctx->zero = tcg_constant_tl(0);
> ctx->virt_inst_excp = false;
> + ctx->decoder = env->decoder;
Up there, in riscv_cpu_finalize_dynamic_decoder(), we assigned a dynamic decoder in
env->decoder. The DisasContext 'decoder' pointer is then assigned to env->decoder in
every tr_init_disas_context() here.
'ctx->decoder' is used in decode_opc(), that receives a CPURISCVState *env pointer as
a parameter.
Am I missing something? It seems we can just remove the ctx->decode pointer altogether
and use env->decoder.
Thanks,
Daniel
> }
>
> static void riscv_tr_tb_start(DisasContextBase *db, CPUState *cpu)
next prev parent reply other threads:[~2024-03-07 19:55 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-03-06 9:33 [PATCH] target/riscv: Implement dynamic establishment of custom decoder Huang Tao
2024-03-07 19:55 ` Daniel Henrique Barboza [this message]
2024-03-07 20:11 ` Richard Henderson
2024-03-07 20:35 ` Richard Henderson
2024-03-08 9:41 ` Huang Tao
2024-03-08 11:56 ` Christoph Müllner
2024-03-07 21:43 ` Daniel Henrique Barboza
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=9b9c1af9-5dca-4270-8dfe-a62223c8cbbb@ventanamicro.com \
--to=dbarboza@ventanamicro.com \
--cc=alistair.francis@wdc.com \
--cc=bin.meng@windriver.com \
--cc=christoph.muellner@vrull.eu \
--cc=eric.huang@linux.alibaba.com \
--cc=liwei1518@gmail.com \
--cc=palmer@dabbelt.com \
--cc=qemu-devel@nongnu.org \
--cc=qemu-riscv@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=zhiwei_liu@linux.alibaba.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).