* RFC: target/riscv: Add support for Zacas extension
@ 2023-06-02 12:16 Rob Bradford
2023-06-02 12:16 ` [PATCH 1/2] target/riscv: Add Zacas ISA extension support Rob Bradford
2023-06-02 12:16 ` [PATCH 2/2] disas/riscv: Add support for amocas.{w,d,q} instructions Rob Bradford
0 siblings, 2 replies; 8+ messages in thread
From: Rob Bradford @ 2023-06-02 12:16 UTC (permalink / raw)
To: qemu-devel; +Cc: qemu-riscv
The Zacas[1] extension is a proposed unprivileged ISA extension for
adding support for atomic compare-and-swap. Since this extension is not
yet frozen (although no significant changes are expected) these patches
are RFC/informational.
Cheers,
Rob
[1] - https://github.com/riscv/riscv-zacas/blob/main/zacas.adoc
^ permalink raw reply [flat|nested] 8+ messages in thread
* [PATCH 1/2] target/riscv: Add Zacas ISA extension support
2023-06-02 12:16 RFC: target/riscv: Add support for Zacas extension Rob Bradford
@ 2023-06-02 12:16 ` Rob Bradford
2023-06-02 15:04 ` Weiwei Li
` (2 more replies)
2023-06-02 12:16 ` [PATCH 2/2] disas/riscv: Add support for amocas.{w,d,q} instructions Rob Bradford
1 sibling, 3 replies; 8+ messages in thread
From: Rob Bradford @ 2023-06-02 12:16 UTC (permalink / raw)
To: qemu-devel
Cc: qemu-riscv, Rob Bradford, Palmer Dabbelt, Alistair Francis,
Bin Meng, Weiwei Li, Daniel Henrique Barboza, Liu Zhiwei
This commit adds support for the the amocas.{w,d,q} instructions behind
a new property to enable that instruction.
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
---
target/riscv/cpu.c | 4 +
target/riscv/cpu.h | 1 +
target/riscv/insn32.decode | 5 +
target/riscv/insn_trans/trans_rvzacas.c.inc | 146 ++++++++++++++++++++
target/riscv/translate.c | 1 +
5 files changed, 157 insertions(+)
create mode 100644 target/riscv/insn_trans/trans_rvzacas.c.inc
diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
index db0875fb43..e99833eb4a 100644
--- a/target/riscv/cpu.c
+++ b/target/riscv/cpu.c
@@ -82,6 +82,7 @@ static const struct isa_ext_data isa_edata_arr[] = {
ISA_EXT_DATA_ENTRY(zicsr, PRIV_VERSION_1_10_0, ext_icsr),
ISA_EXT_DATA_ENTRY(zifencei, PRIV_VERSION_1_10_0, ext_ifencei),
ISA_EXT_DATA_ENTRY(zihintpause, PRIV_VERSION_1_10_0, ext_zihintpause),
+ ISA_EXT_DATA_ENTRY(zacas, PRIV_VERSION_1_12_0, ext_zacas),
ISA_EXT_DATA_ENTRY(zawrs, PRIV_VERSION_1_12_0, ext_zawrs),
ISA_EXT_DATA_ENTRY(zfh, PRIV_VERSION_1_11_0, ext_zfh),
ISA_EXT_DATA_ENTRY(zfhmin, PRIV_VERSION_1_11_0, ext_zfhmin),
@@ -1604,6 +1605,9 @@ static Property riscv_cpu_extensions[] = {
DEFINE_PROP_BOOL("x-zvfh", RISCVCPU, cfg.ext_zvfh, false),
DEFINE_PROP_BOOL("x-zvfhmin", RISCVCPU, cfg.ext_zvfhmin, false),
+ /* Atomic CAS (Zacas) */
+ DEFINE_PROP_BOOL("x-zacas", RISCVCPU, cfg.ext_zacas, false),
+
DEFINE_PROP_END_OF_LIST(),
};
diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
index de7e43126a..ac4d9e8e9c 100644
--- a/target/riscv/cpu.h
+++ b/target/riscv/cpu.h
@@ -436,6 +436,7 @@ struct RISCVCPUConfig {
bool ext_smaia;
bool ext_ssaia;
bool ext_sscofpmf;
+ bool ext_zacas;
bool rvv_ta_all_1s;
bool rvv_ma_all_1s;
diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode
index 73d5d1b045..97d17ee520 100644
--- a/target/riscv/insn32.decode
+++ b/target/riscv/insn32.decode
@@ -255,6 +255,11 @@ amomax_d 10100 . . ..... ..... 011 ..... 0101111 @atom_st
amominu_d 11000 . . ..... ..... 011 ..... 0101111 @atom_st
amomaxu_d 11100 . . ..... ..... 011 ..... 0101111 @atom_st
+# *** Zacas Extension
+amocas_w 00101 . . ..... ..... 010 ..... 0101111 @atom_st
+amocas_d 00101 . . ..... ..... 011 ..... 0101111 @atom_st
+amocas_q 00101 . . ..... ..... 100 ..... 0101111 @atom_st
+
# *** RV32F Standard Extension ***
flw ............ ..... 010 ..... 0000111 @i
fsw ....... ..... ..... 010 ..... 0100111 @s
diff --git a/target/riscv/insn_trans/trans_rvzacas.c.inc b/target/riscv/insn_trans/trans_rvzacas.c.inc
new file mode 100644
index 0000000000..3f1b58ee8a
--- /dev/null
+++ b/target/riscv/insn_trans/trans_rvzacas.c.inc
@@ -0,0 +1,146 @@
+/*
+ * RISC-V translation routines for Zacas extension
+ *
+ * Copyright (c) 2023 Rivos Inc
+ *
+ * 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/>.
+ */
+
+#define REQUIRE_ZACAS(ctx) do { \
+ if (!ctx->cfg_ptr->ext_zacas) { \
+ return false; \
+ } \
+} while (0)
+
+static bool amocas_tl(DisasContext *ctx, arg_atomic *a, MemOp mop)
+{
+ TCGv retv = dest_gpr(ctx, a->rd);
+ TCGv addr = get_address(ctx, a->rs1, 0);
+ TCGv newv = get_gpr(ctx, a->rs2, EXT_ZERO);
+ TCGv cmpv = get_gpr(ctx, a->rd, EXT_ZERO);
+
+ decode_save_opc(ctx);
+ tcg_gen_atomic_cmpxchg_tl(retv, addr, cmpv, newv, ctx->mem_idx, mop);
+ gen_set_gpr(ctx, a->rd, retv);
+
+ return true;
+}
+
+#if TARGET_LONG_BITS == 32
+static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
+{
+ REQUIRE_ZACAS(ctx);
+ return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
+}
+#else
+static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
+{
+ REQUIRE_64BIT(ctx);
+ REQUIRE_ZACAS(ctx);
+ ctx->ol = MXL_RV32;
+ return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
+}
+#endif
+
+#if TARGET_LONG_BITS == 32
+static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
+{
+ TCGv_i64 retv, newv, cmpv;
+ TCGv_i32 cmpv_l, cmpv_h, newv_l, newv_h;
+ TCGv addr;
+
+ REQUIRE_ZACAS(ctx);
+
+ if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
+ return false;
+ }
+
+ addr = get_address(ctx, a->rs1, 0);
+ cmpv_l = get_gpr(ctx, a->rd, 0);
+ cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd + 1, 0);
+ cmpv = tcg_temp_new_i64();
+ tcg_gen_concat_i32_i64(cmpv, cmpv_l, cmpv_h);
+ newv_l = get_gpr(ctx, a->rs2, 0);
+ newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rs2 + 1, 0);
+ newv = tcg_temp_new_i64();
+ tcg_gen_concat_i32_i64(newv, newv_l, newv_h);
+ retv = tcg_temp_new_i64();
+
+ decode_save_opc(ctx);
+ tcg_gen_atomic_cmpxchg_i64(retv, addr, cmpv, newv, ctx->mem_idx,
+ MO_TESQ | MO_ALIGN);
+
+ if (a->rd != 0) {
+ TCGv_i32 retv_l = tcg_temp_new_i32();
+ TCGv_i32 retv_h = tcg_temp_new_i32();
+ tcg_gen_extr_i64_i32(retv_l, retv_h, retv);
+ gen_set_gpr(ctx, a->rd, retv_l);
+ gen_set_gpr(ctx, a->rd + 1, retv_h);
+ }
+
+ return true;
+}
+#else
+static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
+{
+ REQUIRE_64BIT(ctx);
+ REQUIRE_ZACAS(ctx);
+ return amocas_tl(ctx, a, MO_TESQ | MO_ALIGN);
+}
+#endif
+
+#if TARGET_LONG_BITS == 32
+static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
+{
+ return false;
+}
+#else
+static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
+{
+ TCGv_i128 retv, newv, cmpv;
+ TCGv_i64 cmpv_l, cmpv_h, newv_l, newv_h;
+ TCGv addr;
+
+ REQUIRE_64BIT(ctx);
+ REQUIRE_ZACAS(ctx);
+
+ if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
+ return false;
+ }
+
+ addr = get_address(ctx, a->rs1, 0);
+ cmpv_l = get_gpr(ctx, a->rd, 0);
+ cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd + 1, 0);
+ cmpv = tcg_temp_new_i128();
+ tcg_gen_concat_i64_i128(cmpv, cmpv_l, cmpv_h);
+ newv_l = get_gpr(ctx, a->rs2, 0);
+ newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rs2 + 1, 0);
+ newv = tcg_temp_new_i128();
+ tcg_gen_concat_i64_i128(newv, newv_l, newv_h);
+ retv = tcg_temp_new_i128();
+
+ decode_save_opc(ctx);
+ tcg_gen_atomic_cmpxchg_i128(retv, addr, cmpv, newv, ctx->mem_idx,
+ MO_TEUO | MO_ALIGN);
+
+ if (a->rd != 0) {
+ TCGv_i64 retv_l = tcg_temp_new_i64();
+ TCGv_i64 retv_h = tcg_temp_new_i64();
+ tcg_gen_extr_i128_i64(retv_l, retv_h, retv);
+ gen_set_gpr(ctx, a->rd, retv_l);
+ gen_set_gpr(ctx, a->rd + 1, retv_h);
+ }
+
+ return true;
+}
+#endif
diff --git a/target/riscv/translate.c b/target/riscv/translate.c
index 928da0d3f0..55438f5ebf 100644
--- a/target/riscv/translate.c
+++ b/target/riscv/translate.c
@@ -1074,6 +1074,7 @@ static uint32_t opcode_at(DisasContextBase *dcbase, target_ulong pc)
#include "insn_trans/trans_rvv.c.inc"
#include "insn_trans/trans_rvb.c.inc"
#include "insn_trans/trans_rvzicond.c.inc"
+#include "insn_trans/trans_rvzacas.c.inc"
#include "insn_trans/trans_rvzawrs.c.inc"
#include "insn_trans/trans_rvzicbo.c.inc"
#include "insn_trans/trans_rvzfh.c.inc"
--
2.40.1
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [PATCH 2/2] disas/riscv: Add support for amocas.{w,d,q} instructions
2023-06-02 12:16 RFC: target/riscv: Add support for Zacas extension Rob Bradford
2023-06-02 12:16 ` [PATCH 1/2] target/riscv: Add Zacas ISA extension support Rob Bradford
@ 2023-06-02 12:16 ` Rob Bradford
2023-06-02 12:44 ` [PATCH 2/2] disas/riscv: Add support for amocas.{w, d, q} instructions Rob Bradford
1 sibling, 1 reply; 8+ messages in thread
From: Rob Bradford @ 2023-06-02 12:16 UTC (permalink / raw)
To: qemu-devel; +Cc: qemu-riscv, Rob Bradford, Palmer Dabbelt, Alistair Francis
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
---
disas/riscv.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/disas/riscv.c b/disas/riscv.c
index d597161d46..dbed2010f8 100644
--- a/disas/riscv.c
+++ b/disas/riscv.c
@@ -964,6 +964,9 @@ typedef enum {
rv_op_cm_jalt = 788,
rv_op_czero_eqz = 789,
rv_op_czero_nez = 790,
+ rv_op_amocas_w = 791,
+ rv_op_amocas_d = 792,
+ rv_op_amocas_q = 793,
} rv_op;
/* structures */
@@ -2123,6 +2126,9 @@ const rv_opcode_data opcode_data[] = {
{ "cm.jalt", rv_codec_zcmt_jt, rv_fmt_zcmt_index, NULL, 0 },
{ "czero.eqz", rv_codec_r, rv_fmt_rd_rs1_rs2, NULL, 0, 0, 0 },
{ "czero.nez", rv_codec_r, rv_fmt_rd_rs1_rs2, NULL, 0, 0, 0 },
+ { "amoswap.w", rv_codec_r_a, rv_fmt_aqrl_rd_rs2_rs1, NULL, 0, 0, 0 },
+ { "amoswap.d", rv_codec_r_a, rv_fmt_aqrl_rd_rs2_rs1, NULL, 0, 0, 0 },
+ { "amoswap.q", rv_codec_r_a, rv_fmt_aqrl_rd_rs2_rs1, NULL, 0, 0, 0 },
};
/* CSR names */
@@ -2865,6 +2871,9 @@ static void decode_inst_opcode(rv_decode *dec, rv_isa isa)
case 34: op = rv_op_amoxor_w; break;
case 35: op = rv_op_amoxor_d; break;
case 36: op = rv_op_amoxor_q; break;
+ case 42: op = rv_op_amocas_w; break;
+ case 43: op = rv_op_amocas_d; break;
+ case 44: op = rv_op_amocas_q; break;
case 66: op = rv_op_amoor_w; break;
case 67: op = rv_op_amoor_d; break;
case 68: op = rv_op_amoor_q; break;
--
2.40.1
^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: [PATCH 2/2] disas/riscv: Add support for amocas.{w, d, q} instructions
2023-06-02 12:16 ` [PATCH 2/2] disas/riscv: Add support for amocas.{w,d,q} instructions Rob Bradford
@ 2023-06-02 12:44 ` Rob Bradford
0 siblings, 0 replies; 8+ messages in thread
From: Rob Bradford @ 2023-06-02 12:44 UTC (permalink / raw)
To: qemu-devel; +Cc: qemu-riscv, Palmer Dabbelt, Alistair Francis
On Fri, 2 Jun 2023 at 13:17, Rob Bradford <rbradford@rivosinc.com> wrote:
> /* structures */
> @@ -2123,6 +2126,9 @@ const rv_opcode_data opcode_data[] = {
> { "cm.jalt", rv_codec_zcmt_jt, rv_fmt_zcmt_index, NULL, 0 },
> { "czero.eqz", rv_codec_r, rv_fmt_rd_rs1_rs2, NULL, 0, 0, 0 },
> { "czero.nez", rv_codec_r, rv_fmt_rd_rs1_rs2, NULL, 0, 0, 0 },
> + { "amoswap.w", rv_codec_r_a, rv_fmt_aqrl_rd_rs2_rs1, NULL, 0, 0, 0 },
> + { "amoswap.d", rv_codec_r_a, rv_fmt_aqrl_rd_rs2_rs1, NULL, 0, 0, 0 },
> + { "amoswap.q", rv_codec_r_a, rv_fmt_aqrl_rd_rs2_rs1, NULL, 0, 0, 0 },
> };
There is a copy & paste error here. s/amoswap/amocas/.
My local version is fixed already.
Cheers,
Rob
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 1/2] target/riscv: Add Zacas ISA extension support
2023-06-02 12:16 ` [PATCH 1/2] target/riscv: Add Zacas ISA extension support Rob Bradford
@ 2023-06-02 15:04 ` Weiwei Li
2023-06-20 15:22 ` Rob Bradford
2023-06-02 15:08 ` Weiwei Li
2023-06-03 3:28 ` Richard Henderson
2 siblings, 1 reply; 8+ messages in thread
From: Weiwei Li @ 2023-06-02 15:04 UTC (permalink / raw)
To: Rob Bradford, qemu-devel
Cc: qemu-riscv, Palmer Dabbelt, Alistair Francis, Bin Meng,
Weiwei Li, Daniel Henrique Barboza, Liu Zhiwei
On 2023/6/2 20:16, Rob Bradford wrote:
> This commit adds support for the the amocas.{w,d,q} instructions behind
> a new property to enable that instruction.
>
> Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
> ---
I also implemented an initial version for this extension without any
tests a few days ago.
You can find it in https://github.com/plctlab/plct-qemu/tree/plct-zacas-dev.
Hope it can help you.
> target/riscv/cpu.c | 4 +
> target/riscv/cpu.h | 1 +
> target/riscv/insn32.decode | 5 +
> target/riscv/insn_trans/trans_rvzacas.c.inc | 146 ++++++++++++++++++++
> target/riscv/translate.c | 1 +
> 5 files changed, 157 insertions(+)
> create mode 100644 target/riscv/insn_trans/trans_rvzacas.c.inc
> diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
> index db0875fb43..e99833eb4a 100644
> --- a/target/riscv/cpu.c
> +++ b/target/riscv/cpu.c
> @@ -82,6 +82,7 @@ static const struct isa_ext_data isa_edata_arr[] = {
> ISA_EXT_DATA_ENTRY(zicsr, PRIV_VERSION_1_10_0, ext_icsr),
> ISA_EXT_DATA_ENTRY(zifencei, PRIV_VERSION_1_10_0, ext_ifencei),
> ISA_EXT_DATA_ENTRY(zihintpause, PRIV_VERSION_1_10_0, ext_zihintpause),
> + ISA_EXT_DATA_ENTRY(zacas, PRIV_VERSION_1_12_0, ext_zacas),
> ISA_EXT_DATA_ENTRY(zawrs, PRIV_VERSION_1_12_0, ext_zawrs),
> ISA_EXT_DATA_ENTRY(zfh, PRIV_VERSION_1_11_0, ext_zfh),
> ISA_EXT_DATA_ENTRY(zfhmin, PRIV_VERSION_1_11_0, ext_zfhmin),
> @@ -1604,6 +1605,9 @@ static Property riscv_cpu_extensions[] = {
> DEFINE_PROP_BOOL("x-zvfh", RISCVCPU, cfg.ext_zvfh, false),
> DEFINE_PROP_BOOL("x-zvfhmin", RISCVCPU, cfg.ext_zvfhmin, false),
>
> + /* Atomic CAS (Zacas) */
> + DEFINE_PROP_BOOL("x-zacas", RISCVCPU, cfg.ext_zacas, false),
> +
> DEFINE_PROP_END_OF_LIST(),
> };
>
> diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
> index de7e43126a..ac4d9e8e9c 100644
> --- a/target/riscv/cpu.h
> +++ b/target/riscv/cpu.h
> @@ -436,6 +436,7 @@ struct RISCVCPUConfig {
> bool ext_smaia;
> bool ext_ssaia;
> bool ext_sscofpmf;
> + bool ext_zacas;
> bool rvv_ta_all_1s;
> bool rvv_ma_all_1s;
>
> diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode
> index 73d5d1b045..97d17ee520 100644
> --- a/target/riscv/insn32.decode
> +++ b/target/riscv/insn32.decode
> @@ -255,6 +255,11 @@ amomax_d 10100 . . ..... ..... 011 ..... 0101111 @atom_st
> amominu_d 11000 . . ..... ..... 011 ..... 0101111 @atom_st
> amomaxu_d 11100 . . ..... ..... 011 ..... 0101111 @atom_st
>
> +# *** Zacas Extension
> +amocas_w 00101 . . ..... ..... 010 ..... 0101111 @atom_st
> +amocas_d 00101 . . ..... ..... 011 ..... 0101111 @atom_st
> +amocas_q 00101 . . ..... ..... 100 ..... 0101111 @atom_st
> +
> # *** RV32F Standard Extension ***
> flw ............ ..... 010 ..... 0000111 @i
> fsw ....... ..... ..... 010 ..... 0100111 @s
> diff --git a/target/riscv/insn_trans/trans_rvzacas.c.inc b/target/riscv/insn_trans/trans_rvzacas.c.inc
> new file mode 100644
> index 0000000000..3f1b58ee8a
> --- /dev/null
> +++ b/target/riscv/insn_trans/trans_rvzacas.c.inc
> @@ -0,0 +1,146 @@
> +/*
> + * RISC-V translation routines for Zacas extension
> + *
> + * Copyright (c) 2023 Rivos Inc
> + *
> + * 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/>.
> + */
> +
> +#define REQUIRE_ZACAS(ctx) do { \
> + if (!ctx->cfg_ptr->ext_zacas) { \
> + return false; \
> + } \
> +} while (0)
> +
> +static bool amocas_tl(DisasContext *ctx, arg_atomic *a, MemOp mop)
> +{
> + TCGv retv = dest_gpr(ctx, a->rd);
> + TCGv addr = get_address(ctx, a->rs1, 0);
> + TCGv newv = get_gpr(ctx, a->rs2, EXT_ZERO);
> + TCGv cmpv = get_gpr(ctx, a->rd, EXT_ZERO);
> +
> + decode_save_opc(ctx);
> + tcg_gen_atomic_cmpxchg_tl(retv, addr, cmpv, newv, ctx->mem_idx, mop);
> + gen_set_gpr(ctx, a->rd, retv);
> +
> + return true;
> +}
> +
> +#if TARGET_LONG_BITS == 32
> +static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
> +{
> + REQUIRE_ZACAS(ctx);
> + return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
> +}
> +#else
> +static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
> +{
> + REQUIRE_64BIT(ctx);
> + REQUIRE_ZACAS(ctx);
> + ctx->ol = MXL_RV32;
> + return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
> +}
> +#endif
This seems unnecessary to use different trans function for RV32 and RV64.
tcg_gen_atomic_cmpxchg_tl itself will do the actually cmp and xchg
operation based on size in memop
instead of the target long size.
> +
> +#if TARGET_LONG_BITS == 32
We commonly use TARGET_RISCV32 instead of TARGET_LONG_BITS in similar
cases.
> +static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
> +{
> + TCGv_i64 retv, newv, cmpv;
> + TCGv_i32 cmpv_l, cmpv_h, newv_l, newv_h;
> + TCGv addr;
> +
> + REQUIRE_ZACAS(ctx);
> +
> + if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
> + return false;
> + }
> +
> + addr = get_address(ctx, a->rs1, 0);
> + cmpv_l = get_gpr(ctx, a->rd, 0);
It's better to use EXT_NONE instead of 0 in this and following cases.
> + cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd + 1, 0);
We can use ctx->zero for get_gpr(ctx, 0, 0) directly here.
> + cmpv = tcg_temp_new_i64();
> + tcg_gen_concat_i32_i64(cmpv, cmpv_l, cmpv_h);
> + newv_l = get_gpr(ctx, a->rs2, 0);
> + newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rs2 + 1, 0);
> + newv = tcg_temp_new_i64();
> + tcg_gen_concat_i32_i64(newv, newv_l, newv_h);
> + retv = tcg_temp_new_i64();
> +
> + decode_save_opc(ctx);
> + tcg_gen_atomic_cmpxchg_i64(retv, addr, cmpv, newv, ctx->mem_idx,
> + MO_TESQ | MO_ALIGN);
> +
> + if (a->rd != 0) {
> + TCGv_i32 retv_l = tcg_temp_new_i32();
> + TCGv_i32 retv_h = tcg_temp_new_i32();
> + tcg_gen_extr_i64_i32(retv_l, retv_h, retv);
> + gen_set_gpr(ctx, a->rd, retv_l);
> + gen_set_gpr(ctx, a->rd + 1, retv_h);
> + }
> +
> + return true;
> +}
> +#else
> +static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
> +{
> + REQUIRE_64BIT(ctx);
This seems not correct. amocas_d can work on RV32, so I think it can
also work on RV64 with xl = 32
> + REQUIRE_ZACAS(ctx);
> + return amocas_tl(ctx, a, MO_TESQ | MO_ALIGN);
> +}
> +#endif
> +
> +#if TARGET_LONG_BITS == 32
> +static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
> +{
> + return false;
> +}
This case is unnecessary.
We can just add REQUIRE_64BIT(ctx) in following trans_amocas_q
> +#else
> +static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
> +{
> + TCGv_i128 retv, newv, cmpv;
> + TCGv_i64 cmpv_l, cmpv_h, newv_l, newv_h;
> + TCGv addr;
> +
> + REQUIRE_64BIT(ctx);
> + REQUIRE_ZACAS(ctx);
> +
> + if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
> + return false;
> + }
> +
> + addr = get_address(ctx, a->rs1, 0);
> + cmpv_l = get_gpr(ctx, a->rd, 0);
> + cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd + 1, 0);
> + cmpv = tcg_temp_new_i128();
> + tcg_gen_concat_i64_i128(cmpv, cmpv_l, cmpv_h);
> + newv_l = get_gpr(ctx, a->rs2, 0);
> + newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rs2 + 1, 0);
> + newv = tcg_temp_new_i128();
> + tcg_gen_concat_i64_i128(newv, newv_l, newv_h);
> + retv = tcg_temp_new_i128();
> +
> + decode_save_opc(ctx);
> + tcg_gen_atomic_cmpxchg_i128(retv, addr, cmpv, newv, ctx->mem_idx,
> + MO_TEUO | MO_ALIGN);
> +
> + if (a->rd != 0) {
> + TCGv_i64 retv_l = tcg_temp_new_i64();
> + TCGv_i64 retv_h = tcg_temp_new_i64();
> + tcg_gen_extr_i128_i64(retv_l, retv_h, retv);
> + gen_set_gpr(ctx, a->rd, retv_l);
> + gen_set_gpr(ctx, a->rd + 1, retv_h);
> + }
> +
> + return true;
> +}
> +#endif
> diff --git a/target/riscv/translate.c b/target/riscv/translate.c
> index 928da0d3f0..55438f5ebf 100644
> --- a/target/riscv/translate.c
> +++ b/target/riscv/translate.c
> @@ -1074,6 +1074,7 @@ static uint32_t opcode_at(DisasContextBase *dcbase, target_ulong pc)
> #include "insn_trans/trans_rvv.c.inc"
> #include "insn_trans/trans_rvb.c.inc"
> #include "insn_trans/trans_rvzicond.c.inc"
> +#include "insn_trans/trans_rvzacas.c.inc"
> #include "insn_trans/trans_rvzawrs.c.inc"
> #include "insn_trans/trans_rvzicbo.c.inc"
> #include "insn_trans/trans_rvzfh.c.inc"
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 1/2] target/riscv: Add Zacas ISA extension support
2023-06-02 12:16 ` [PATCH 1/2] target/riscv: Add Zacas ISA extension support Rob Bradford
2023-06-02 15:04 ` Weiwei Li
@ 2023-06-02 15:08 ` Weiwei Li
2023-06-03 3:28 ` Richard Henderson
2 siblings, 0 replies; 8+ messages in thread
From: Weiwei Li @ 2023-06-02 15:08 UTC (permalink / raw)
To: Rob Bradford, qemu-devel
Cc: qemu-riscv, Palmer Dabbelt, Alistair Francis, Bin Meng,
Weiwei Li, Daniel Henrique Barboza, Liu Zhiwei
On 2023/6/2 20:16, Rob Bradford wrote:
> This commit adds support for the the amocas.{w,d,q} instructions behind
> a new property to enable that instruction.
>
> Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
> ---
> target/riscv/cpu.c | 4 +
> target/riscv/cpu.h | 1 +
> target/riscv/insn32.decode | 5 +
> target/riscv/insn_trans/trans_rvzacas.c.inc | 146 ++++++++++++++++++++
> target/riscv/translate.c | 1 +
> 5 files changed, 157 insertions(+)
> create mode 100644 target/riscv/insn_trans/trans_rvzacas.c.inc
>
> diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
> index db0875fb43..e99833eb4a 100644
> --- a/target/riscv/cpu.c
> +++ b/target/riscv/cpu.c
> @@ -82,6 +82,7 @@ static const struct isa_ext_data isa_edata_arr[] = {
> ISA_EXT_DATA_ENTRY(zicsr, PRIV_VERSION_1_10_0, ext_icsr),
> ISA_EXT_DATA_ENTRY(zifencei, PRIV_VERSION_1_10_0, ext_ifencei),
> ISA_EXT_DATA_ENTRY(zihintpause, PRIV_VERSION_1_10_0, ext_zihintpause),
> + ISA_EXT_DATA_ENTRY(zacas, PRIV_VERSION_1_12_0, ext_zacas),
> ISA_EXT_DATA_ENTRY(zawrs, PRIV_VERSION_1_12_0, ext_zawrs),
> ISA_EXT_DATA_ENTRY(zfh, PRIV_VERSION_1_11_0, ext_zfh),
> ISA_EXT_DATA_ENTRY(zfhmin, PRIV_VERSION_1_11_0, ext_zfhmin),
> @@ -1604,6 +1605,9 @@ static Property riscv_cpu_extensions[] = {
> DEFINE_PROP_BOOL("x-zvfh", RISCVCPU, cfg.ext_zvfh, false),
> DEFINE_PROP_BOOL("x-zvfhmin", RISCVCPU, cfg.ext_zvfhmin, false),
>
> + /* Atomic CAS (Zacas) */
> + DEFINE_PROP_BOOL("x-zacas", RISCVCPU, cfg.ext_zacas, false),
> +
> DEFINE_PROP_END_OF_LIST(),
> };
>
> diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
> index de7e43126a..ac4d9e8e9c 100644
> --- a/target/riscv/cpu.h
> +++ b/target/riscv/cpu.h
> @@ -436,6 +436,7 @@ struct RISCVCPUConfig {
> bool ext_smaia;
> bool ext_ssaia;
> bool ext_sscofpmf;
> + bool ext_zacas;
> bool rvv_ta_all_1s;
> bool rvv_ma_all_1s;
>
> diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode
> index 73d5d1b045..97d17ee520 100644
> --- a/target/riscv/insn32.decode
> +++ b/target/riscv/insn32.decode
> @@ -255,6 +255,11 @@ amomax_d 10100 . . ..... ..... 011 ..... 0101111 @atom_st
> amominu_d 11000 . . ..... ..... 011 ..... 0101111 @atom_st
> amomaxu_d 11100 . . ..... ..... 011 ..... 0101111 @atom_st
>
> +# *** Zacas Extension
> +amocas_w 00101 . . ..... ..... 010 ..... 0101111 @atom_st
> +amocas_d 00101 . . ..... ..... 011 ..... 0101111 @atom_st
> +amocas_q 00101 . . ..... ..... 100 ..... 0101111 @atom_st
> +
> # *** RV32F Standard Extension ***
> flw ............ ..... 010 ..... 0000111 @i
> fsw ....... ..... ..... 010 ..... 0100111 @s
> diff --git a/target/riscv/insn_trans/trans_rvzacas.c.inc b/target/riscv/insn_trans/trans_rvzacas.c.inc
> new file mode 100644
> index 0000000000..3f1b58ee8a
> --- /dev/null
> +++ b/target/riscv/insn_trans/trans_rvzacas.c.inc
> @@ -0,0 +1,146 @@
> +/*
> + * RISC-V translation routines for Zacas extension
> + *
> + * Copyright (c) 2023 Rivos Inc
> + *
> + * 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/>.
> + */
> +
> +#define REQUIRE_ZACAS(ctx) do { \
> + if (!ctx->cfg_ptr->ext_zacas) { \
> + return false; \
> + } \
> +} while (0)
> +
> +static bool amocas_tl(DisasContext *ctx, arg_atomic *a, MemOp mop)
> +{
> + TCGv retv = dest_gpr(ctx, a->rd);
> + TCGv addr = get_address(ctx, a->rs1, 0);
> + TCGv newv = get_gpr(ctx, a->rs2, EXT_ZERO);
> + TCGv cmpv = get_gpr(ctx, a->rd, EXT_ZERO);
> +
> + decode_save_opc(ctx);
> + tcg_gen_atomic_cmpxchg_tl(retv, addr, cmpv, newv, ctx->mem_idx, mop);
> + gen_set_gpr(ctx, a->rd, retv);
> +
> + return true;
> +}
> +
> +#if TARGET_LONG_BITS == 32
> +static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
> +{
> + REQUIRE_ZACAS(ctx);
> + return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
> +}
> +#else
> +static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
> +{
> + REQUIRE_64BIT(ctx);
> + REQUIRE_ZACAS(ctx);
> + ctx->ol = MXL_RV32;
> + return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
> +}
> +#endif
> +
> +#if TARGET_LONG_BITS == 32
> +static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
> +{
> + TCGv_i64 retv, newv, cmpv;
> + TCGv_i32 cmpv_l, cmpv_h, newv_l, newv_h;
> + TCGv addr;
> +
> + REQUIRE_ZACAS(ctx);
> +
> + if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
> + return false;
> + }
> +
> + addr = get_address(ctx, a->rs1, 0);
> + cmpv_l = get_gpr(ctx, a->rd, 0);
> + cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd + 1, 0);
> + cmpv = tcg_temp_new_i64();
> + tcg_gen_concat_i32_i64(cmpv, cmpv_l, cmpv_h);
> + newv_l = get_gpr(ctx, a->rs2, 0);
> + newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rs2 + 1, 0);
> + newv = tcg_temp_new_i64();
> + tcg_gen_concat_i32_i64(newv, newv_l, newv_h);
> + retv = tcg_temp_new_i64();
> +
> + decode_save_opc(ctx);
> + tcg_gen_atomic_cmpxchg_i64(retv, addr, cmpv, newv, ctx->mem_idx,
> + MO_TESQ | MO_ALIGN);
> +
> + if (a->rd != 0) {
> + TCGv_i32 retv_l = tcg_temp_new_i32();
> + TCGv_i32 retv_h = tcg_temp_new_i32();
> + tcg_gen_extr_i64_i32(retv_l, retv_h, retv);
> + gen_set_gpr(ctx, a->rd, retv_l);
> + gen_set_gpr(ctx, a->rd + 1, retv_h);
> + }
> +
> + return true;
> +}
> +#else
> +static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
> +{
> + REQUIRE_64BIT(ctx);
> + REQUIRE_ZACAS(ctx);
> + return amocas_tl(ctx, a, MO_TESQ | MO_ALIGN);
> +}
> +#endif
> +
> +#if TARGET_LONG_BITS == 32
> +static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
> +{
> + return false;
> +}
> +#else
> +static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
> +{
> + TCGv_i128 retv, newv, cmpv;
> + TCGv_i64 cmpv_l, cmpv_h, newv_l, newv_h;
> + TCGv addr;
> +
> + REQUIRE_64BIT(ctx);
> + REQUIRE_ZACAS(ctx);
> +
> + if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
> + return false;
> + }
> +
> + addr = get_address(ctx, a->rs1, 0);
> + cmpv_l = get_gpr(ctx, a->rd, 0);
> + cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd + 1, 0);
> + cmpv = tcg_temp_new_i128();
> + tcg_gen_concat_i64_i128(cmpv, cmpv_l, cmpv_h);
> + newv_l = get_gpr(ctx, a->rs2, 0);
> + newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rs2 + 1, 0);
> + newv = tcg_temp_new_i128();
> + tcg_gen_concat_i64_i128(newv, newv_l, newv_h);
> + retv = tcg_temp_new_i128();
> +
> + decode_save_opc(ctx);
> + tcg_gen_atomic_cmpxchg_i128(retv, addr, cmpv, newv, ctx->mem_idx,
> + MO_TEUO | MO_ALIGN);
> +
> + if (a->rd != 0) {
> + TCGv_i64 retv_l = tcg_temp_new_i64();
> + TCGv_i64 retv_h = tcg_temp_new_i64();
> + tcg_gen_extr_i128_i64(retv_l, retv_h, retv);
> + gen_set_gpr(ctx, a->rd, retv_l);
> + gen_set_gpr(ctx, a->rd + 1, retv_h);
> + }
> +
> + return true;
> +}
> +#endif
> diff --git a/target/riscv/translate.c b/target/riscv/translate.c
> index 928da0d3f0..55438f5ebf 100644
> --- a/target/riscv/translate.c
> +++ b/target/riscv/translate.c
> @@ -1074,6 +1074,7 @@ static uint32_t opcode_at(DisasContextBase *dcbase, target_ulong pc)
> #include "insn_trans/trans_rvv.c.inc"
> #include "insn_trans/trans_rvb.c.inc"
> #include "insn_trans/trans_rvzicond.c.inc"
> +#include "insn_trans/trans_rvzacas.c.inc"
> #include "insn_trans/trans_rvzawrs.c.inc"
> #include "insn_trans/trans_rvzicbo.c.inc"
> #include "insn_trans/trans_rvzfh.c.inc"
It seems lack check on "Zacas requires A"
Regards,
Weiwei Li
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 1/2] target/riscv: Add Zacas ISA extension support
2023-06-02 12:16 ` [PATCH 1/2] target/riscv: Add Zacas ISA extension support Rob Bradford
2023-06-02 15:04 ` Weiwei Li
2023-06-02 15:08 ` Weiwei Li
@ 2023-06-03 3:28 ` Richard Henderson
2 siblings, 0 replies; 8+ messages in thread
From: Richard Henderson @ 2023-06-03 3:28 UTC (permalink / raw)
To: Rob Bradford, qemu-devel
Cc: qemu-riscv, Palmer Dabbelt, Alistair Francis, Bin Meng,
Weiwei Li, Daniel Henrique Barboza, Liu Zhiwei
On 6/2/23 05:16, Rob Bradford wrote:
> +#if TARGET_LONG_BITS == 32
> +static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
You need to eliminate all of the ifdefs, because we can switch a 64-bit cpu into 32-bit
mode -- get_xl(ctx) shows which mode we are in.
r~
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 1/2] target/riscv: Add Zacas ISA extension support
2023-06-02 15:04 ` Weiwei Li
@ 2023-06-20 15:22 ` Rob Bradford
0 siblings, 0 replies; 8+ messages in thread
From: Rob Bradford @ 2023-06-20 15:22 UTC (permalink / raw)
To: Weiwei Li, qemu-devel
Cc: qemu-riscv, Palmer Dabbelt, Alistair Francis, Bin Meng,
Daniel Henrique Barboza, Liu Zhiwei
On Fri, 2023-06-02 at 23:04 +0800, Weiwei Li wrote:
>
> On 2023/6/2 20:16, Rob Bradford wrote:
> > This commit adds support for the the amocas.{w,d,q} instructions
> > behind
> > a new property to enable that instruction.
> >
> > Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
> > ---
>
> I also implemented an initial version for this extension without any
> tests a few days ago.
>
> You can find it in
> https://github.com/plctlab/plct-qemu/tree/plct-zacas-dev.
Hi, thank you for sharing this! Sorry for the delay in response. I had
a chance to test your version and found a small issue with the .Q
implementation when I tested it.
Here is the fix:
diff --git a/target/riscv/insn_trans/trans_rvzacas.c.inc
b/target/riscv/insn_trans/trans_rvzacas.c.inc
index c063079c6a..e556f48da8 100644
--- a/target/riscv/insn_trans/trans_rvzacas.c.inc
+++ b/target/riscv/insn_trans/trans_rvzacas.c.inc
@@ -129,11 +129,12 @@ static bool trans_amocas_q(DisasContext *ctx,
arg_amocas_q *a)
TCGv_i128 src2 = tcg_temp_new_i128();
TCGv_i64 src2l = get_gpr(ctx, a->rs2, EXT_NONE);
TCGv_i64 src2h = get_gpr(ctx, a->rs2 == 0 ? 0 : a->rs2 + 1,
EXT_NONE);
- TCGv_i64 destl = get_gpr(ctx, a->rs2, EXT_NONE);
- TCGv_i64 desth = get_gpr(ctx, a->rs2 == 0 ? 0 : a->rs2 + 1,
EXT_NONE);
+ TCGv_i64 destl = get_gpr(ctx, a->rd, EXT_NONE);
+ TCGv_i64 desth = get_gpr(ctx, a->rd == 0 ? 0 : a->rd + 1,
EXT_NONE);
tcg_gen_concat_i64_i128(src2, src2l, src2h);
tcg_gen_concat_i64_i128(dest, destl, desth);
+ decode_save_opc(ctx);
tcg_gen_atomic_cmpxchg_i128(dest, src1, dest, src2, ctx->mem_idx,
(MO_ALIGN | MO_TEUO));
Since you already have an implementation ready i'm happy for your
version to be the one included.
>
> Hope it can help you.
>
> > target/riscv/cpu.c | 4 +
> > target/riscv/cpu.h | 1 +
> > target/riscv/insn32.decode | 5 +
> > target/riscv/insn_trans/trans_rvzacas.c.inc | 146
> > ++++++++++++++++++++
> > target/riscv/translate.c | 1 +
> > 5 files changed, 157 insertions(+)
> > create mode 100644 target/riscv/insn_trans/trans_rvzacas.c.inc
> > diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
> > index db0875fb43..e99833eb4a 100644
> > --- a/target/riscv/cpu.c
> > +++ b/target/riscv/cpu.c
> > @@ -82,6 +82,7 @@ static const struct isa_ext_data isa_edata_arr[]
> > = {
> > ISA_EXT_DATA_ENTRY(zicsr, PRIV_VERSION_1_10_0, ext_icsr),
> > ISA_EXT_DATA_ENTRY(zifencei, PRIV_VERSION_1_10_0,
> > ext_ifencei),
> > ISA_EXT_DATA_ENTRY(zihintpause, PRIV_VERSION_1_10_0,
> > ext_zihintpause),
> > + ISA_EXT_DATA_ENTRY(zacas, PRIV_VERSION_1_12_0, ext_zacas),
> > ISA_EXT_DATA_ENTRY(zawrs, PRIV_VERSION_1_12_0, ext_zawrs),
> > ISA_EXT_DATA_ENTRY(zfh, PRIV_VERSION_1_11_0, ext_zfh),
> > ISA_EXT_DATA_ENTRY(zfhmin, PRIV_VERSION_1_11_0, ext_zfhmin),
> > @@ -1604,6 +1605,9 @@ static Property riscv_cpu_extensions[] = {
> > DEFINE_PROP_BOOL("x-zvfh", RISCVCPU, cfg.ext_zvfh, false),
> > DEFINE_PROP_BOOL("x-zvfhmin", RISCVCPU, cfg.ext_zvfhmin,
> > false),
> >
> > + /* Atomic CAS (Zacas) */
> > + DEFINE_PROP_BOOL("x-zacas", RISCVCPU, cfg.ext_zacas, false),
> > +
> > DEFINE_PROP_END_OF_LIST(),
> > };
> >
> > diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
> > index de7e43126a..ac4d9e8e9c 100644
> > --- a/target/riscv/cpu.h
> > +++ b/target/riscv/cpu.h
> > @@ -436,6 +436,7 @@ struct RISCVCPUConfig {
> > bool ext_smaia;
> > bool ext_ssaia;
> > bool ext_sscofpmf;
> > + bool ext_zacas;
> > bool rvv_ta_all_1s;
> > bool rvv_ma_all_1s;
> >
> > diff --git a/target/riscv/insn32.decode
> > b/target/riscv/insn32.decode
> > index 73d5d1b045..97d17ee520 100644
> > --- a/target/riscv/insn32.decode
> > +++ b/target/riscv/insn32.decode
> > @@ -255,6 +255,11 @@ amomax_d 10100 . . ..... ..... 011 .....
> > 0101111 @atom_st
> > amominu_d 11000 . . ..... ..... 011 ..... 0101111 @atom_st
> > amomaxu_d 11100 . . ..... ..... 011 ..... 0101111 @atom_st
> >
> > +# *** Zacas Extension
> > +amocas_w 00101 . . ..... ..... 010 ..... 0101111 @atom_st
> > +amocas_d 00101 . . ..... ..... 011 ..... 0101111 @atom_st
> > +amocas_q 00101 . . ..... ..... 100 ..... 0101111 @atom_st
> > +
> > # *** RV32F Standard Extension ***
> > flw ............ ..... 010 ..... 0000111 @i
> > fsw ....... ..... ..... 010 ..... 0100111 @s
> > diff --git a/target/riscv/insn_trans/trans_rvzacas.c.inc
> > b/target/riscv/insn_trans/trans_rvzacas.c.inc
> > new file mode 100644
> > index 0000000000..3f1b58ee8a
> > --- /dev/null
> > +++ b/target/riscv/insn_trans/trans_rvzacas.c.inc
> > @@ -0,0 +1,146 @@
> > +/*
> > + * RISC-V translation routines for Zacas extension
> > + *
> > + * Copyright (c) 2023 Rivos Inc
> > + *
> > + * 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/>.
> > + */
> > +
> > +#define REQUIRE_ZACAS(ctx) do { \
> > + if (!ctx->cfg_ptr->ext_zacas) { \
> > + return false; \
> > + } \
> > +} while (0)
> > +
> > +static bool amocas_tl(DisasContext *ctx, arg_atomic *a, MemOp mop)
> > +{
> > + TCGv retv = dest_gpr(ctx, a->rd);
> > + TCGv addr = get_address(ctx, a->rs1, 0);
> > + TCGv newv = get_gpr(ctx, a->rs2, EXT_ZERO);
> > + TCGv cmpv = get_gpr(ctx, a->rd, EXT_ZERO);
> > +
> > + decode_save_opc(ctx);
> > + tcg_gen_atomic_cmpxchg_tl(retv, addr, cmpv, newv, ctx-
> > >mem_idx, mop);
> > + gen_set_gpr(ctx, a->rd, retv);
> > +
> > + return true;
> > +}
> > +
> > +#if TARGET_LONG_BITS == 32
> > +static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
> > +{
> > + REQUIRE_ZACAS(ctx);
> > + return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
> > +}
> > +#else
> > +static bool trans_amocas_w(DisasContext *ctx, arg_amocas_w *a)
> > +{
> > + REQUIRE_64BIT(ctx);
> > + REQUIRE_ZACAS(ctx);
> > + ctx->ol = MXL_RV32;
> > + return amocas_tl(ctx, a, MO_TESL | MO_ALIGN);
> > +}
> > +#endif
>
> This seems unnecessary to use different trans function for RV32 and
> RV64.
>
> tcg_gen_atomic_cmpxchg_tl itself will do the actually cmp and xchg
> operation based on size in memop
>
> instead of the target long size.
Thank you - I can see from your version how this would work.
>
> > +
> > +#if TARGET_LONG_BITS == 32
> We commonly use TARGET_RISCV32 instead of TARGET_LONG_BITS in similar
> cases.
Noted - thank you. Richard in another thread also pointed out my
version didn't handle dynamic xlen correctly.
> > +static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
> > +{
> > + TCGv_i64 retv, newv, cmpv;
> > + TCGv_i32 cmpv_l, cmpv_h, newv_l, newv_h;
> > + TCGv addr;
> > +
> > + REQUIRE_ZACAS(ctx);
> > +
> > + if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
> > + return false;
> > + }
> > +
> > + addr = get_address(ctx, a->rs1, 0);
> > + cmpv_l = get_gpr(ctx, a->rd, 0);
> It's better to use EXT_NONE instead of 0 in this and following cases.
Noted!
> > + cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd
> > + 1, 0);
> We can use ctx->zero for get_gpr(ctx, 0, 0) directly here.
Thanks!
> > + cmpv = tcg_temp_new_i64();
> > + tcg_gen_concat_i32_i64(cmpv, cmpv_l, cmpv_h);
> > + newv_l = get_gpr(ctx, a->rs2, 0);
> > + newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a-
> > >rs2 + 1, 0);
> > + newv = tcg_temp_new_i64();
> > + tcg_gen_concat_i32_i64(newv, newv_l, newv_h);
> > + retv = tcg_temp_new_i64();
> > +
> > + decode_save_opc(ctx);
> > + tcg_gen_atomic_cmpxchg_i64(retv, addr, cmpv, newv, ctx-
> > >mem_idx,
> > + MO_TESQ | MO_ALIGN);
> > +
> > + if (a->rd != 0) {
> > + TCGv_i32 retv_l = tcg_temp_new_i32();
> > + TCGv_i32 retv_h = tcg_temp_new_i32();
> > + tcg_gen_extr_i64_i32(retv_l, retv_h, retv);
> > + gen_set_gpr(ctx, a->rd, retv_l);
> > + gen_set_gpr(ctx, a->rd + 1, retv_h);
> > + }
> > +
> > + return true;
> > +}
> > +#else
> > +static bool trans_amocas_d(DisasContext *ctx, arg_amocas_w *a)
> > +{
> > + REQUIRE_64BIT(ctx);
> This seems not correct. amocas_d can work on RV32, so I think it can
> also work on RV64 with xl = 32
> > + REQUIRE_ZACAS(ctx);
> > + return amocas_tl(ctx, a, MO_TESQ | MO_ALIGN);
> > +}
> > +#endif
> > +
> > +#if TARGET_LONG_BITS == 32
> > +static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
> > +{
> > + return false;
> > +}
>
> This case is unnecessary.
>
> We can just add REQUIRE_64BIT(ctx) in following trans_amocas_q
>
Thank you for the review - this was my first TCG patch and i've learnt
a lot from that.
Cheers,
Rob
> > +#else
> > +static bool trans_amocas_q(DisasContext *ctx, arg_amocas_w *a)
> > +{
> > + TCGv_i128 retv, newv, cmpv;
> > + TCGv_i64 cmpv_l, cmpv_h, newv_l, newv_h;
> > + TCGv addr;
> > +
> > + REQUIRE_64BIT(ctx);
> > + REQUIRE_ZACAS(ctx);
> > +
> > + if (a->rd % 2 == 1 || a->rs2 % 2 == 1) {
> > + return false;
> > + }
> > +
> > + addr = get_address(ctx, a->rs1, 0);
> > + cmpv_l = get_gpr(ctx, a->rd, 0);
> > + cmpv_h = a->rd == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a->rd
> > + 1, 0);
> > + cmpv = tcg_temp_new_i128();
> > + tcg_gen_concat_i64_i128(cmpv, cmpv_l, cmpv_h);
> > + newv_l = get_gpr(ctx, a->rs2, 0);
> > + newv_h = a->rs2 == 0 ? get_gpr(ctx, 0, 0) : get_gpr(ctx, a-
> > >rs2 + 1, 0);
> > + newv = tcg_temp_new_i128();
> > + tcg_gen_concat_i64_i128(newv, newv_l, newv_h);
> > + retv = tcg_temp_new_i128();
> > +
> > + decode_save_opc(ctx);
> > + tcg_gen_atomic_cmpxchg_i128(retv, addr, cmpv, newv, ctx-
> > >mem_idx,
> > + MO_TEUO | MO_ALIGN);
> > +
> > + if (a->rd != 0) {
> > + TCGv_i64 retv_l = tcg_temp_new_i64();
> > + TCGv_i64 retv_h = tcg_temp_new_i64();
> > + tcg_gen_extr_i128_i64(retv_l, retv_h, retv);
> > + gen_set_gpr(ctx, a->rd, retv_l);
> > + gen_set_gpr(ctx, a->rd + 1, retv_h);
> > + }
> > +
> > + return true;
> > +}
> > +#endif
> > diff --git a/target/riscv/translate.c b/target/riscv/translate.c
> > index 928da0d3f0..55438f5ebf 100644
> > --- a/target/riscv/translate.c
> > +++ b/target/riscv/translate.c
> > @@ -1074,6 +1074,7 @@ static uint32_t opcode_at(DisasContextBase
> > *dcbase, target_ulong pc)
> > #include "insn_trans/trans_rvv.c.inc"
> > #include "insn_trans/trans_rvb.c.inc"
> > #include "insn_trans/trans_rvzicond.c.inc"
> > +#include "insn_trans/trans_rvzacas.c.inc"
> > #include "insn_trans/trans_rvzawrs.c.inc"
> > #include "insn_trans/trans_rvzicbo.c.inc"
> > #include "insn_trans/trans_rvzfh.c.inc"
>
^ permalink raw reply related [flat|nested] 8+ messages in thread
end of thread, other threads:[~2023-06-20 15:23 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-06-02 12:16 RFC: target/riscv: Add support for Zacas extension Rob Bradford
2023-06-02 12:16 ` [PATCH 1/2] target/riscv: Add Zacas ISA extension support Rob Bradford
2023-06-02 15:04 ` Weiwei Li
2023-06-20 15:22 ` Rob Bradford
2023-06-02 15:08 ` Weiwei Li
2023-06-03 3:28 ` Richard Henderson
2023-06-02 12:16 ` [PATCH 2/2] disas/riscv: Add support for amocas.{w,d,q} instructions Rob Bradford
2023-06-02 12:44 ` [PATCH 2/2] disas/riscv: Add support for amocas.{w, d, q} instructions Rob Bradford
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).