[RFC PATCH 33/39] target/riscv: Add vghmac.vv decoding, translation and execution support

[Lawrence Hunter]

Co-authored-by: Nazar Kazakov <nazar.kazakov@codethink.co.uk>
Signed-off-by: Nazar Kazakov <nazar.kazakov@codethink.co.uk>
Signed-off-by: Lawrence Hunter <lawrence.hunter@codethink.co.uk>
---
 target/riscv/helper.h                      |  2 +
 target/riscv/insn32.decode                 |  3 ++
 target/riscv/insn_trans/trans_rvzvkg.c.inc |  8 ++++
 target/riscv/translate.c                   |  1 +
 target/riscv/vcrypto_helper.c              | 45 ++++++++++++++++++++++
 5 files changed, 59 insertions(+)
 create mode 100644 target/riscv/insn_trans/trans_rvzvkg.c.inc

diff --git a/target/riscv/helper.h b/target/riscv/helper.h
index a82103ead9..6272294d50 100644
--- a/target/riscv/helper.h
+++ b/target/riscv/helper.h
@@ -1199,3 +1199,5 @@ DEF_HELPER_5(vsha2cl_vv, void, ptr, ptr, ptr, env, i32)
 
 DEF_HELPER_5(vsm3me_vv, void, ptr, ptr, ptr, env, i32)
 DEF_HELPER_5(vsm3c_vi, void, ptr, ptr, i32, env, i32)
+
+DEF_HELPER_5(vghmac_vv, void, ptr, ptr, ptr, env, i32)
diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode
index 4a50114e92..ff044f8288 100644
--- a/target/riscv/insn32.decode
+++ b/target/riscv/insn32.decode
@@ -930,3 +930,6 @@ vsha2cl_vv      101111 1 ..... ..... 010 ..... 1110111 
@r_vm_1
 # *** RV64 Zvksh vector crypto extensions ***
 vsm3me_vv       100000 1 ..... ..... 010 ..... 1110111 @r_vm_1
 vsm3c_vi        101011 1 ..... ..... 010 ..... 1110111 @r_vm_1
+
+# *** RV64 Zvkg vector crypto extension ***
+vghmac_vv       101100 1 ..... ..... 010 ..... 1110111 @r_vm_1
diff --git a/target/riscv/insn_trans/trans_rvzvkg.c.inc 
b/target/riscv/insn_trans/trans_rvzvkg.c.inc
new file mode 100644
index 0000000000..576f5d39c4
--- /dev/null
+++ b/target/riscv/insn_trans/trans_rvzvkg.c.inc
@@ -0,0 +1,8 @@
+static bool vghmac_check(DisasContext *s, arg_rmrr *a)
+{
+    return opivv_check(s, a) &&
+           s->vstart % 4 == 0 &&
+           s->sew == MO_32;
+}
+
+GEN_VV_UNMASKED_TRANS(vghmac_vv, vghmac_check)
diff --git a/target/riscv/translate.c b/target/riscv/translate.c
index 9ca2cec23a..0bc1c9db65 100644
--- a/target/riscv/translate.c
+++ b/target/riscv/translate.c
@@ -1067,6 +1067,7 @@ static uint32_t opcode_at(DisasContextBase *dcbase, 
target_ulong pc)
 #include "insn_trans/trans_rvzvkns.c.inc"
 #include "insn_trans/trans_rvzvknh.c.inc"
 #include "insn_trans/trans_rvzvksh.c.inc"
+#include "insn_trans/trans_rvzvkg.c.inc"
 #include "insn_trans/trans_privileged.c.inc"
 #include "insn_trans/trans_svinval.c.inc"
 #include "insn_trans/trans_xventanacondops.c.inc"
diff --git a/target/riscv/vcrypto_helper.c b/target/riscv/vcrypto_helper.c
index 478e652c9b..a309ac3f03 100644
--- a/target/riscv/vcrypto_helper.c
+++ b/target/riscv/vcrypto_helper.c
@@ -827,3 +827,48 @@ void HELPER(vsm3c_vi)(void *vd_vptr, void *vs2_vptr, 
uint32_t uimm,
     vext_set_elems_1s(vd_vptr, vta, env->vl * esz, total_elems * esz);
     env->vstart = 0;
 }
+
+void HELPER(vghmac_vv)(void *vd_vptr, void *vs1_vptr, void *vs2_vptr,
+                       CPURISCVState *env, uint32_t desc)
+{
+    uint64_t *vd = vd_vptr;
+    uint64_t *vs1 = vs1_vptr;
+    uint64_t *vs2 = vs2_vptr;
+    uint32_t vta = vext_vta(desc);
+    uint32_t total_elems = vext_get_total_elems(env, desc, 4);
+
+    if (env->vl % 4 != 0) {
+        riscv_raise_exception(env, RISCV_EXCP_ILLEGAL_INST, GETPC());
+    }
+
+    for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
+        __uint128_t Y = reverse_bits_byte_8(vd[i * 2 + 0]) |
+                        ((__uint128_t)reverse_bits_byte_8(vd[i * 2 + 1]) << 
64);
+        __uint128_t H = reverse_bits_byte_8(vs1[i * 2 + 0]) |
+                        ((__uint128_t)reverse_bits_byte_8(vs1[i * 2 + 1]) << 
64);
+        __uint128_t X = vs2[i * 2 + 0] | ((__uint128_t)vs2[i * 2 + 1] << 64);
+        __uint128_t Z = 0;
+
+        for (uint j = 0; j < 128; j++) {
+            if ((Y >> j) & 1) {
+                Z ^= H;
+            }
+            bool reduce = ((H >> 127) & 1);
+            H = H << 1;
+            if (reduce) {
+                H ^= 0x87;
+            }
+        }
+
+        Z = reverse_bits_byte_8(Z & UINT64_MAX) |
+            ((__uint128_t)reverse_bits_byte_8((Z >> 64) & UINT64_MAX) << 64);
+
+        Z = Z ^ X;
+
+        vd[i * 2 + 0] = Z & UINT64_MAX;
+        vd[i * 2 + 1] = (Z >> 64) & UINT64_MAX;
+    }
+    /* set tail elements to 1s */
+    vext_set_elems_1s(vd, vta, env->vl * 4, total_elems * 4);
+    env->vstart = 0;
+}
-- 
2.39.1