diff mbox series

[v3,09/14] RISC-V: Adding T-Head MemIdx extension

Message ID 20230124195945.181842-10-christoph.muellner@vrull.eu (mailing list archive)
State New, archived
Headers show
Series Add support for the T-Head vendor extensions | expand

Commit Message

Christoph Müllner Jan. 24, 2023, 7:59 p.m. UTC 
From: Christoph Müllner <christoph.muellner@vrull.eu>

This patch adds support for the T-Head MemIdx instructions.
The patch uses the T-Head specific decoder and translation.

Co-developed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Christoph Müllner <christoph.muellner@vrull.eu>
---
Changes in v2:
- Add ISA_EXT_DATA_ENTRY()
- Use single decoder for XThead extensions
- Avoid signed-bitfield-extraction by using signed immediate field imm5
- Use get_address() to calculate addresses
- Introduce helper get_th_address_indexed for rs1+(rs2<<imm2) calculation
- Introduce get_address_indexed() for register offsets (like get_address())

 target/riscv/cpu.c                         |   2 +
 target/riscv/cpu.h                         |   1 +
 target/riscv/insn_trans/trans_xthead.c.inc | 377 +++++++++++++++++++++
 target/riscv/translate.c                   |  21 +-
 target/riscv/xthead.decode                 |  54 +++
 5 files changed, 454 insertions(+), 1 deletion(-)

Comments

Richard Henderson Jan. 24, 2023, 9:21 p.m. UTC | #1 
On 1/24/23 09:59, Christoph Muellner wrote:
> +/* XTheadMemIdx */
> +
> +/*
> + * Load with memop from indexed address and add (imm5 << imm2) to rs1.
> + * If !preinc, then the load address is rs1.
> + * If  preinc, then the load address is rs1 + (imm5) << imm2).
> + */
> +static bool gen_load_inc(DisasContext *ctx, arg_th_meminc *a, MemOp memop,
> +                         bool preinc)
> +{
> +    TCGv rd = dest_gpr(ctx, a->rd);
> +    TCGv addr = get_address(ctx, a->rs1, preinc ? a->imm5 << a->imm2 : 0);
> +
> +    tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
> +    addr = get_address(ctx, a->rs1, !preinc ? a->imm5 << a->imm2 : 0);

First, you're leaking the previous 'addr' temporary.
Second, get_address may make modifications to 'addr' which you don't want to write back.
Third, you are not checking for rd != rs1.

I think what you want is

     int imm = a->imm5 << a->imm2;
     TCGv addr = get_address(ctx, a->rs1, preinc ? imm : 0);
     TCGv rd = dest_gpr(ctx, a->rd);
     TCGv rs1 = get_gpr(ctx, a->rs1, EXT_NONE);

     tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
     tcg_gen_addi_tl(rs1, rs1, imm);
     gen_set_gpr(ctx, a->rd, rd);
     gen_set_gpr(ctx, a->rs1, rs1);


r~
LIU Zhiwei Jan. 30, 2023, 9:04 a.m. UTC | #2 
On 2023/1/25 5:21, Richard Henderson wrote:
> On 1/24/23 09:59, Christoph Muellner wrote:
>> +/* XTheadMemIdx */
>> +
>> +/*
>> + * Load with memop from indexed address and add (imm5 << imm2) to rs1.
>> + * If !preinc, then the load address is rs1.
>> + * If  preinc, then the load address is rs1 + (imm5) << imm2).
>> + */
>> +static bool gen_load_inc(DisasContext *ctx, arg_th_meminc *a, MemOp 
>> memop,
>> +                         bool preinc)
>> +{
>> +    TCGv rd = dest_gpr(ctx, a->rd);
>> +    TCGv addr = get_address(ctx, a->rs1, preinc ? a->imm5 << a->imm2 
>> : 0);
>> +
>> +    tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
>> +    addr = get_address(ctx, a->rs1, !preinc ? a->imm5 << a->imm2 : 0);
>
> First, you're leaking the previous 'addr' temporary.
Although all temps allocated by temp_new() will be freed after the 
instruction translation automatically, we can improve current 
implementation.
> Second, get_address may make modifications to 'addr' which you don't 
> want to write back.
Good catch.
> Third, you are not checking for rd != rs1.
Yes.
>
> I think what you want is
>
>     int imm = a->imm5 << a->imm2;
>     TCGv addr = get_address(ctx, a->rs1, preinc ? imm : 0);
>     TCGv rd = dest_gpr(ctx, a->rd);
>     TCGv rs1 = get_gpr(ctx, a->rs1, EXT_NONE);
>
>     tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
>     tcg_gen_addi_tl(rs1, rs1, imm);
>     gen_set_gpr(ctx, a->rd, rd);
>     gen_set_gpr(ctx, a->rs1, rs1);

Yes, we should write back the 'addr' without modification.

Best Regards,
Zhiwei

>
>
> r~
Christoph Müllner Jan. 31, 2023, 6:01 p.m. UTC | #3 
On Tue, Jan 24, 2023 at 10:21 PM Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> On 1/24/23 09:59, Christoph Muellner wrote:
> > +/* XTheadMemIdx */
> > +
> > +/*
> > + * Load with memop from indexed address and add (imm5 << imm2) to rs1.
> > + * If !preinc, then the load address is rs1.
> > + * If  preinc, then the load address is rs1 + (imm5) << imm2).
> > + */
> > +static bool gen_load_inc(DisasContext *ctx, arg_th_meminc *a, MemOp memop,
> > +                         bool preinc)
> > +{
> > +    TCGv rd = dest_gpr(ctx, a->rd);
> > +    TCGv addr = get_address(ctx, a->rs1, preinc ? a->imm5 << a->imm2 : 0);
> > +
> > +    tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
> > +    addr = get_address(ctx, a->rs1, !preinc ? a->imm5 << a->imm2 : 0);
>
> First, you're leaking the previous 'addr' temporary.

Indeed!
The real question is of course, why we call get_address() twice...
Fixed in v4.

> Second, get_address may make modifications to 'addr' which you don't want to write back.

Fixed in v4.

> Third, you are not checking for rd != rs1.

Fixed in v4.

>
> I think what you want is
>
>      int imm = a->imm5 << a->imm2;
>      TCGv addr = get_address(ctx, a->rs1, preinc ? imm : 0);
>      TCGv rd = dest_gpr(ctx, a->rd);
>      TCGv rs1 = get_gpr(ctx, a->rs1, EXT_NONE);
>
>      tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
>      tcg_gen_addi_tl(rs1, rs1, imm);
>      gen_set_gpr(ctx, a->rd, rd);
>      gen_set_gpr(ctx, a->rs1, rs1);

Yes (plus the check for rd != rs1).

Fixed in v4.

Thank you!

>
>
> r~
diff mbox series

Patch

diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
index e3a10f782c..eb8bbfa436 100644
--- a/target/riscv/cpu.c
+++ b/target/riscv/cpu.c
@@ -115,6 +115,7 @@  static const struct isa_ext_data isa_edata_arr[] = {
     ISA_EXT_DATA_ENTRY(xtheadcmo, true, PRIV_VERSION_1_11_0, ext_xtheadcmo),
     ISA_EXT_DATA_ENTRY(xtheadcondmov, true, PRIV_VERSION_1_11_0, ext_xtheadcondmov),
     ISA_EXT_DATA_ENTRY(xtheadmac, true, PRIV_VERSION_1_11_0, ext_xtheadmac),
+    ISA_EXT_DATA_ENTRY(xtheadmemidx, true, PRIV_VERSION_1_11_0, ext_xtheadmemidx),
     ISA_EXT_DATA_ENTRY(xtheadmempair, true, PRIV_VERSION_1_11_0, ext_xtheadmempair),
     ISA_EXT_DATA_ENTRY(xtheadsync, true, PRIV_VERSION_1_11_0, ext_xtheadsync),
     ISA_EXT_DATA_ENTRY(xventanacondops, true, PRIV_VERSION_1_12_0, ext_XVentanaCondOps),
@@ -1085,6 +1086,7 @@  static Property riscv_cpu_extensions[] = {
     DEFINE_PROP_BOOL("xtheadcmo", RISCVCPU, cfg.ext_xtheadcmo, false),
     DEFINE_PROP_BOOL("xtheadcondmov", RISCVCPU, cfg.ext_xtheadcondmov, false),
     DEFINE_PROP_BOOL("xtheadmac", RISCVCPU, cfg.ext_xtheadmac, false),
+    DEFINE_PROP_BOOL("xtheadmemidx", RISCVCPU, cfg.ext_xtheadmemidx, false),
     DEFINE_PROP_BOOL("xtheadmempair", RISCVCPU, cfg.ext_xtheadmempair, false),
     DEFINE_PROP_BOOL("xtheadsync", RISCVCPU, cfg.ext_xtheadsync, false),
     DEFINE_PROP_BOOL("xventanacondops", RISCVCPU, cfg.ext_XVentanaCondOps, false),
diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
index 4f5f3b2c20..4882b9a9cc 100644
--- a/target/riscv/cpu.h
+++ b/target/riscv/cpu.h
@@ -479,6 +479,7 @@  struct RISCVCPUConfig {
     bool ext_xtheadcmo;
     bool ext_xtheadcondmov;
     bool ext_xtheadmac;
+    bool ext_xtheadmemidx;
     bool ext_xtheadmempair;
     bool ext_xtheadsync;
     bool ext_XVentanaCondOps;
diff --git a/target/riscv/insn_trans/trans_xthead.c.inc b/target/riscv/insn_trans/trans_xthead.c.inc
index 7ab2a7a48e..e41f3be9a6 100644
--- a/target/riscv/insn_trans/trans_xthead.c.inc
+++ b/target/riscv/insn_trans/trans_xthead.c.inc
@@ -52,6 +52,12 @@ 
     }                                            \
 } while (0)
 
+#define REQUIRE_XTHEADMEMIDX(ctx) do {           \
+    if (!ctx->cfg_ptr->ext_xtheadmemidx) {       \
+        return false;                            \
+    }                                            \
+} while (0)
+
 #define REQUIRE_XTHEADMEMPAIR(ctx) do {          \
     if (!ctx->cfg_ptr->ext_xtheadmempair) {      \
         return false;                            \
@@ -64,6 +70,30 @@ 
     }                                            \
 } while (0)
 
+/*
+ * Calculate and return the address for indexed mem operations:
+ * If !zext_offs, then the address is rs1 + (rs2 << imm2).
+ * If  zext_offs, then the address is rs1 + (zext(rs2[31:0]) << imm2).
+ */
+static TCGv get_th_address_indexed(DisasContext *ctx, int rs1, int rs2,
+                                   int imm2, bool zext_offs)
+{
+    TCGv src2 = get_gpr(ctx, rs2, EXT_NONE);
+    TCGv offs = tcg_temp_new();
+
+    if (zext_offs) {
+        tcg_gen_extract_tl(offs, src2, 0, 32);
+        tcg_gen_shli_tl(offs, offs, imm2);
+    } else {
+        tcg_gen_shli_tl(offs, src2, imm2);
+    }
+
+    TCGv addr = get_address_indexed(ctx, rs1, offs);
+
+    tcg_temp_free(offs);
+    return addr;
+}
+
 /* XTheadBa */
 
 /*
@@ -388,6 +418,353 @@  static bool trans_th_mulsw(DisasContext *ctx, arg_th_mulsw *a)
     return gen_th_mac(ctx, a, tcg_gen_sub_tl, NULL);
 }
 
+/* XTheadMemIdx */
+
+/*
+ * Load with memop from indexed address and add (imm5 << imm2) to rs1.
+ * If !preinc, then the load address is rs1.
+ * If  preinc, then the load address is rs1 + (imm5) << imm2).
+ */
+static bool gen_load_inc(DisasContext *ctx, arg_th_meminc *a, MemOp memop,
+                         bool preinc)
+{
+    TCGv rd = dest_gpr(ctx, a->rd);
+    TCGv addr = get_address(ctx, a->rs1, preinc ? a->imm5 << a->imm2 : 0);
+
+    tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
+    addr = get_address(ctx, a->rs1, !preinc ? a->imm5 << a->imm2 : 0);
+    gen_set_gpr(ctx, a->rd, rd);
+    gen_set_gpr(ctx, a->rs1, addr);
+
+    return true;
+}
+
+/*
+ * Store with memop to indexed address and add (imm5 << imm2) to rs1.
+ * If !preinc, then the store address is rs1.
+ * If  preinc, then the store address is rs1 + (imm5) << imm2).
+ */
+static bool gen_store_inc(DisasContext *ctx, arg_th_meminc *a, MemOp memop,
+                          bool preinc)
+{
+    TCGv data = get_gpr(ctx, a->rd, EXT_NONE);
+    TCGv addr = get_address(ctx, a->rs1, preinc ? a->imm5 << a->imm2 : 0);
+
+    tcg_gen_qemu_st_tl(data, addr, ctx->mem_idx, memop);
+    addr = get_address(ctx, a->rs1, !preinc ? a->imm5 << a->imm2 : 0);
+    gen_set_gpr(ctx, a->rs1, addr);
+
+    return true;
+}
+
+static bool trans_th_ldia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_inc(ctx, a, MO_TESQ, false);
+}
+
+static bool trans_th_ldib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_inc(ctx, a, MO_TESQ, true);
+}
+
+static bool trans_th_lwia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_TESL, false);
+}
+
+static bool trans_th_lwib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_TESL, true);
+}
+
+static bool trans_th_lwuia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_inc(ctx, a, MO_TEUL, false);
+}
+
+static bool trans_th_lwuib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_inc(ctx, a, MO_TEUL, true);
+}
+
+static bool trans_th_lhia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_TESW, false);
+}
+
+static bool trans_th_lhib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_TESW, true);
+}
+
+static bool trans_th_lhuia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_TEUW, false);
+}
+
+static bool trans_th_lhuib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_TEUW, true);
+}
+
+static bool trans_th_lbia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_SB, false);
+}
+
+static bool trans_th_lbib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_SB, true);
+}
+
+static bool trans_th_lbuia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_UB, false);
+}
+
+static bool trans_th_lbuib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_inc(ctx, a, MO_UB, true);
+}
+
+static bool trans_th_sdia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_store_inc(ctx, a, MO_TESQ, false);
+}
+
+static bool trans_th_sdib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_store_inc(ctx, a, MO_TESQ, true);
+}
+
+static bool trans_th_swia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_inc(ctx, a, MO_TESL, false);
+}
+
+static bool trans_th_swib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_inc(ctx, a, MO_TESL, true);
+}
+
+static bool trans_th_shia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_inc(ctx, a, MO_TESW, false);
+}
+
+static bool trans_th_shib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_inc(ctx, a, MO_TESW, true);
+}
+
+static bool trans_th_sbia(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_inc(ctx, a, MO_SB, false);
+}
+
+static bool trans_th_sbib(DisasContext *ctx, arg_th_meminc *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_inc(ctx, a, MO_SB, true);
+}
+
+/*
+ * Load with memop from indexed address.
+ * If !zext_offs, then address is rs1 + (rs2 << imm2).
+ * If  zext_offs, then address is rs1 + (zext(rs2[31:0]) << imm2).
+ */
+static bool gen_load_idx(DisasContext *ctx, arg_th_memidx *a, MemOp memop,
+                         bool zext_offs)
+{
+    TCGv rd = dest_gpr(ctx, a->rd);
+    TCGv addr = get_th_address_indexed(ctx, a->rs1, a->rs2, a->imm2, zext_offs);
+
+    tcg_gen_qemu_ld_tl(rd, addr, ctx->mem_idx, memop);
+    gen_set_gpr(ctx, a->rd, rd);
+
+    return true;
+}
+
+/*
+ * Store with memop to indexed address.
+ * If !zext_offs, then address is rs1 + (rs2 << imm2).
+ * If  zext_offs, then address is rs1 + (zext(rs2[31:0]) << imm2).
+ */
+static bool gen_store_idx(DisasContext *ctx, arg_th_memidx *a, MemOp memop,
+                          bool zext_offs)
+{
+    TCGv data = get_gpr(ctx, a->rd, EXT_NONE);
+    TCGv addr = get_th_address_indexed(ctx, a->rs1, a->rs2, a->imm2, zext_offs);
+
+    tcg_gen_qemu_st_tl(data, addr, ctx->mem_idx, memop);
+
+    return true;
+}
+
+static bool trans_th_lrd(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_idx(ctx, a, MO_TESQ, false);
+}
+
+static bool trans_th_lrw(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_TESL, false);
+}
+
+static bool trans_th_lrwu(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_idx(ctx, a, MO_TEUL, false);
+}
+
+static bool trans_th_lrh(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_TESW, false);
+}
+
+static bool trans_th_lrhu(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_TEUW, false);
+}
+
+static bool trans_th_lrb(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_SB, false);
+}
+
+static bool trans_th_lrbu(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_UB, false);
+}
+
+static bool trans_th_srd(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_store_idx(ctx, a, MO_TESQ, false);
+}
+
+static bool trans_th_srw(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_idx(ctx, a, MO_TESL, false);
+}
+
+static bool trans_th_srh(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_idx(ctx, a, MO_TESW, false);
+}
+
+static bool trans_th_srb(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_idx(ctx, a, MO_SB, false);
+}
+static bool trans_th_lurd(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_idx(ctx, a, MO_TESQ, true);
+}
+
+static bool trans_th_lurw(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_TESL, true);
+}
+
+static bool trans_th_lurwu(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_load_idx(ctx, a, MO_TEUL, true);
+}
+
+static bool trans_th_lurh(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_TESW, true);
+}
+
+static bool trans_th_lurhu(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_TEUW, true);
+}
+
+static bool trans_th_lurb(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_SB, true);
+}
+
+static bool trans_th_lurbu(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_load_idx(ctx, a, MO_UB, true);
+}
+
+static bool trans_th_surd(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    REQUIRE_64BIT(ctx);
+    return gen_store_idx(ctx, a, MO_TESQ, true);
+}
+
+static bool trans_th_surw(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_idx(ctx, a, MO_TESL, true);
+}
+
+static bool trans_th_surh(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_idx(ctx, a, MO_TESW, true);
+}
+
+static bool trans_th_surb(DisasContext *ctx, arg_th_memidx *a)
+{
+    REQUIRE_XTHEADMEMIDX(ctx);
+    return gen_store_idx(ctx, a, MO_SB, true);
+}
+
 /* XTheadMemPair */
 
 static bool gen_loadpair_tl(DisasContext *ctx, arg_th_pair *a, MemOp memop,
diff --git a/target/riscv/translate.c b/target/riscv/translate.c
index 27bab07994..2461bcda0f 100644
--- a/target/riscv/translate.c
+++ b/target/riscv/translate.c
@@ -133,7 +133,8 @@  static bool has_xthead_p(DisasContext *ctx  __attribute__((__unused__)))
     return ctx->cfg_ptr->ext_xtheadba || ctx->cfg_ptr->ext_xtheadbb ||
            ctx->cfg_ptr->ext_xtheadbs || ctx->cfg_ptr->ext_xtheadcmo ||
            ctx->cfg_ptr->ext_xtheadcondmov || ctx->cfg_ptr->ext_xtheadmac ||
-           ctx->cfg_ptr->ext_xtheadmempair || ctx->cfg_ptr->ext_xtheadsync;
+           ctx->cfg_ptr->ext_xtheadmemidx || ctx->cfg_ptr->ext_xtheadmempair ||
+           ctx->cfg_ptr->ext_xtheadsync;
 }
 
 #define MATERIALISE_EXT_PREDICATE(ext)  \
@@ -595,6 +596,24 @@  static TCGv get_address(DisasContext *ctx, int rs1, int imm)
     return addr;
 }
 
+/* Compute a canonical address from a register plus reg offset. */
+static TCGv get_address_indexed(DisasContext *ctx, int rs1, TCGv offs)
+{
+    TCGv addr = temp_new(ctx);
+    TCGv src1 = get_gpr(ctx, rs1, EXT_NONE);
+
+    tcg_gen_add_tl(addr, src1, offs);
+    if (ctx->pm_mask_enabled) {
+        tcg_gen_andc_tl(addr, addr, pm_mask);
+    } else if (get_xl(ctx) == MXL_RV32) {
+        tcg_gen_ext32u_tl(addr, addr);
+    }
+    if (ctx->pm_base_enabled) {
+        tcg_gen_or_tl(addr, addr, pm_base);
+    }
+    return addr;
+}
+
 #ifndef CONFIG_USER_ONLY
 /* The states of mstatus_fs are:
  * 0 = disabled, 1 = initial, 2 = clean, 3 = dirty
diff --git a/target/riscv/xthead.decode b/target/riscv/xthead.decode
index ff2a83b56d..69e40f22dc 100644
--- a/target/riscv/xthead.decode
+++ b/target/riscv/xthead.decode
@@ -17,8 +17,10 @@ 
 %rd2       20:5
 %rs2       20:5
 %sh5       20:5
+%imm5      20:s5
 %sh6       20:6
 %sh2       25:2
+%imm2      25:2
 
 # Argument sets
 &r         rd rs1 rs2                               !extern
@@ -26,6 +28,8 @@ 
 &shift     shamt rs1 rd                             !extern
 &th_bfext  msb lsb rs1 rd
 &th_pair   rd1 rs rd2 sh2
+&th_memidx rd rs1 rs2 imm2
+&th_meminc rd rs1 imm5 imm2
 
 # Formats
 @sfence_vm  ....... ..... .....   ... ..... ....... %rs1
@@ -36,6 +40,8 @@ 
 @sh5        ....... ..... .....  ... ..... .......  &shift  shamt=%sh5      %rs1 %rd
 @sh6        ...... ...... .....  ... ..... .......  &shift shamt=%sh6 %rs1 %rd
 @th_pair    ..... .. ..... ..... ... ..... .......  &th_pair %rd1 %rs %rd2 %sh2
+@th_memidx  ..... .. ..... ..... ... ..... .......  &th_memidx %rd %rs1 %rs2 %imm2
+@th_meminc  ..... .. ..... ..... ... ..... .......  &th_meminc %rd %rs1 %imm5 %imm2
 
 # XTheadBa
 # Instead of defining a new encoding, we simply use the decoder to
@@ -102,6 +108,54 @@  th_muls          00100 01 ..... ..... 001 ..... 0001011 @r
 th_mulsh         00101 01 ..... ..... 001 ..... 0001011 @r
 th_mulsw         00100 11 ..... ..... 001 ..... 0001011 @r
 
+# XTheadMemIdx
+th_ldia          01111 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_ldib          01101 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lwia          01011 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lwib          01001 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lwuia         11011 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lwuib         11001 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lhia          00111 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lhib          00101 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lhuia         10111 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lhuib         10101 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lbia          00011 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lbib          00001 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lbuia         10011 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_lbuib         10001 .. ..... ..... 100 ..... 0001011 @th_meminc
+th_sdia          01111 .. ..... ..... 101 ..... 0001011 @th_meminc
+th_sdib          01101 .. ..... ..... 101 ..... 0001011 @th_meminc
+th_swia          01011 .. ..... ..... 101 ..... 0001011 @th_meminc
+th_swib          01001 .. ..... ..... 101 ..... 0001011 @th_meminc
+th_shia          00111 .. ..... ..... 101 ..... 0001011 @th_meminc
+th_shib          00101 .. ..... ..... 101 ..... 0001011 @th_meminc
+th_sbia          00011 .. ..... ..... 101 ..... 0001011 @th_meminc
+th_sbib          00001 .. ..... ..... 101 ..... 0001011 @th_meminc
+
+th_lrd           01100 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lrw           01000 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lrwu          11000 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lrh           00100 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lrhu          10100 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lrb           00000 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lrbu          10000 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_srd           01100 .. ..... ..... 101 ..... 0001011 @th_memidx
+th_srw           01000 .. ..... ..... 101 ..... 0001011 @th_memidx
+th_srh           00100 .. ..... ..... 101 ..... 0001011 @th_memidx
+th_srb           00000 .. ..... ..... 101 ..... 0001011 @th_memidx
+
+th_lurd          01110 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lurw          01010 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lurwu         11010 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lurh          00110 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lurhu         10110 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lurb          00010 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_lurbu         10010 .. ..... ..... 100 ..... 0001011 @th_memidx
+th_surd          01110 .. ..... ..... 101 ..... 0001011 @th_memidx
+th_surw          01010 .. ..... ..... 101 ..... 0001011 @th_memidx
+th_surh          00110 .. ..... ..... 101 ..... 0001011 @th_memidx
+th_surb          00010 .. ..... ..... 101 ..... 0001011 @th_memidx
+
 # XTheadMemPair
 th_ldd           11111 .. ..... ..... 100 ..... 0001011 @th_pair
 th_lwd           11100 .. ..... ..... 100 ..... 0001011 @th_pair