@@ -925,6 +925,7 @@ static Property riscv_cpu_extensions[] = {
DEFINE_PROP_BOOL("xtheadbs", RISCVCPU, cfg.ext_xtheadbs, false),
DEFINE_PROP_BOOL("xtheadcmo", RISCVCPU, cfg.ext_xtheadcmo, false),
DEFINE_PROP_BOOL("xtheadcondmov", RISCVCPU, cfg.ext_xtheadcondmov, false),
+ DEFINE_PROP_BOOL("xtheadfmemidx", RISCVCPU, cfg.ext_xtheadfmemidx, 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),
@@ -445,6 +445,7 @@ struct RISCVCPUConfig {
bool ext_xtheadbs;
bool ext_xtheadcmo;
bool ext_xtheadcondmov;
+ bool ext_xtheadfmemidx;
bool ext_xtheadmac;
bool ext_xtheadmemidx;
bool ext_xtheadmempair;
@@ -751,3 +751,124 @@ static bool trans_th_surb(DisasContext *ctx, arg_th_memidx *a)
return gen_store_idx(ctx, a, MO_SB, true);
}
+/*
+ * Load 64-bit float from indexed address.
+ * If !zero_extend_offset, then address is rs1 + (rs2 << imm2).
+ * If zero_extend_offset, then address is rs1 + (zext(rs2[31:0]) << imm2).
+ */
+static bool gen_fload_idx(DisasContext *ctx, arg_th_fmemidx *a, MemOp memop,
+ bool zero_extend_offset)
+{
+ TCGv_i64 rd = cpu_fpr[a->rd];
+ TCGv base = get_gpr(ctx, a->rs1, EXT_NONE);
+ TCGv offs = get_gpr(ctx, a->rs2, EXT_NONE);
+ TCGv addr = tcg_temp_new();
+
+ if (zero_extend_offset) {
+ tcg_gen_extract_tl(addr, offs, 0, 32);
+ } else {
+ tcg_gen_mov_tl(addr, offs);
+ }
+ tcg_gen_shli_tl(addr, addr, a->imm2);
+ tcg_gen_add_tl(addr, base, addr);
+
+ if (get_xl(ctx) == MXL_RV32) {
+ tcg_gen_ext32u_tl(addr, addr);
+ }
+
+ tcg_gen_qemu_ld_i64(rd, addr, ctx->mem_idx, memop);
+ if ((memop & MO_SIZE) == MO_32) {
+ gen_nanbox_s(rd, rd);
+ }
+
+ mark_fs_dirty(ctx);
+ tcg_temp_free(addr);
+ return true;
+}
+
+/*
+ * Store 64-bit float to indexed address.
+ * If !zero_extend_offset, then address is rs1 + (rs2 << imm2).
+ * If zero_extend_offset, then address is rs1 + (zext(rs2[31:0]) << imm2).
+ */
+static bool gen_fstore_idx(DisasContext *ctx, arg_th_fmemidx *a, MemOp memop,
+ bool zero_extend_offset)
+{
+ TCGv_i64 rd = cpu_fpr[a->rd];
+ TCGv base = get_gpr(ctx, a->rs1, EXT_NONE);
+ TCGv offs = get_gpr(ctx, a->rs2, EXT_NONE);
+ TCGv addr = tcg_temp_new();
+
+ if (zero_extend_offset) {
+ tcg_gen_extract_tl(addr, offs, 0, 32);
+ } else {
+ tcg_gen_mov_tl(addr, offs);
+ }
+ tcg_gen_shli_tl(addr, addr, a->imm2);
+ tcg_gen_add_tl(addr, base, addr);
+
+ if (get_xl(ctx) == MXL_RV32) {
+ tcg_gen_ext32u_tl(addr, addr);
+ }
+
+ tcg_gen_qemu_st_i64(rd, addr, ctx->mem_idx, memop);
+
+ tcg_temp_free(addr);
+ return true;
+}
+
+static bool trans_th_flrd(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVD);
+ return gen_fload_idx(ctx, a, MO_TEUQ, false);
+}
+
+static bool trans_th_flrw(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVF);
+ return gen_fload_idx(ctx, a, MO_TEUL, false);
+}
+
+static bool trans_th_flurd(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVD);
+ return gen_fload_idx(ctx, a, MO_TEUQ, true);
+}
+
+static bool trans_th_flurw(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVF);
+ return gen_fload_idx(ctx, a, MO_TEUL, true);
+}
+
+static bool trans_th_fsrd(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVD);
+ return gen_fstore_idx(ctx, a, MO_TEUQ, false);
+}
+
+static bool trans_th_fsrw(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVF);
+ return gen_fstore_idx(ctx, a, MO_TEUL, false);
+}
+
+static bool trans_th_fsurd(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVD);
+ return gen_fstore_idx(ctx, a, MO_TEUQ, true);
+}
+
+static bool trans_th_fsurw(DisasContext *ctx, arg_th_fmemidx *a)
+{
+ REQUIRE_FPU;
+ REQUIRE_EXT(ctx, RVF);
+ return gen_fstore_idx(ctx, a, MO_TEUL, true);
+}
@@ -7,6 +7,7 @@ gen = [
decodetree.process('xtheadbs.decode', extra_args: '--static-decode=decode_xtheadbs'),
decodetree.process('xtheadcmo.decode', extra_args: '--static-decode=decode_xtheadcmo'),
decodetree.process('xtheadcondmov.decode', extra_args: '--static-decode=decode_xtheadcondmov'),
+ decodetree.process('xtheadfmemidx.decode', extra_args: '--static-decode=decode_xtheadfmemidx'),
decodetree.process('xtheadmac.decode', extra_args: '--static-decode=decode_xtheadmac'),
decodetree.process('xtheadmemidx.decode', extra_args: '--static-decode=decode_xtheadmemidx'),
decodetree.process('xtheadmempair.decode', extra_args: '--static-decode=decode_xtheadmempair'),
@@ -137,6 +137,7 @@ MATERIALISE_EXT_PREDICATE(xtheadbb)
MATERIALISE_EXT_PREDICATE(xtheadbs)
MATERIALISE_EXT_PREDICATE(xtheadcmo)
MATERIALISE_EXT_PREDICATE(xtheadcondmov);
+MATERIALISE_EXT_PREDICATE(xtheadfmemidx);
MATERIALISE_EXT_PREDICATE(xtheadmac);
MATERIALISE_EXT_PREDICATE(xtheadmemidx);
MATERIALISE_EXT_PREDICATE(xtheadmempair);
@@ -732,6 +733,7 @@ static int ex_rvc_shifti(DisasContext *ctx, int imm)
#include "decode-xtheadbs.c.inc"
#include "decode-xtheadcmo.c.inc"
#include "decode-xtheadcondmov.c.inc"
+#include "decode-xtheadfmemidx.c.inc"
#include "decode-xtheadmac.c.inc"
#include "decode-xtheadmemidx.c.inc"
#include "decode-xtheadmempair.c.inc"
@@ -1061,6 +1063,7 @@ static void decode_opc(CPURISCVState *env, DisasContext *ctx, uint16_t opcode)
{ has_xtheadbs_p, decode_xtheadbs },
{ has_xtheadcmo_p, decode_xtheadcmo },
{ has_xtheadcondmov_p, decode_xtheadcondmov },
+ { has_xtheadfmemidx_p, decode_xtheadfmemidx },
{ has_xtheadmac_p, decode_xtheadmac },
{ has_xtheadmemidx_p, decode_xtheadmemidx },
{ has_xtheadmempair_p, decode_xtheadmempair },
new file mode 100644
@@ -0,0 +1,34 @@
+#
+# RISC-V instruction decode for the XTheadMemIdx extension
+#
+# Copyright (c) 2022 Christoph Muellner, christoph.muellner@vrull.eu
+#
+# SPDX-License-Identifier: LGPL-2.1-or-later
+#
+# The XTheadFMemIdx extension provides floating-point memory operations.
+#
+# It is documented in
+# https://github.com/T-head-Semi/thead-extension-spec/releases/download/2.0.0/xthead-2022-09-05-2.0.0.pdf
+
+# Fields
+%imm2 25:2
+%rs2 20:5
+%rs1 15:5
+%rd 7:5
+
+# Argument sets
+&th_fmemidx rd rs1 rs2 imm2
+
+# Formats
+@th_fmemidx ..... .. ..... ..... ... ..... ....... &th_fmemidx %rd %rs1 %rs2 %imm2
+
+# Instructions
+th_flrd 01100 .. ..... ..... 110 ..... 0001011 @th_fmemidx
+th_flrw 01000 .. ..... ..... 110 ..... 0001011 @th_fmemidx
+th_flurd 01110 .. ..... ..... 110 ..... 0001011 @th_fmemidx
+th_flurw 01010 .. ..... ..... 110 ..... 0001011 @th_fmemidx
+
+th_fsrd 01100 .. ..... ..... 111 ..... 0001011 @th_fmemidx
+th_fsrw 01000 .. ..... ..... 111 ..... 0001011 @th_fmemidx
+th_fsurd 01110 .. ..... ..... 111 ..... 0001011 @th_fmemidx
+th_fsurw 01010 .. ..... ..... 111 ..... 0001011 @th_fmemidx