[PATCH v6 05/14] tcg/riscv: Implement vector mov/dup{m/i}

[Richard Henderson <richard.henderson@linaro.org>]

From: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>

Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20241007025700.47259-5-zhiwei_liu@linux.alibaba.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/riscv/tcg-target.c.inc | 76 +++++++++++++++++++++++++++++++++++++-
 1 file changed, 74 insertions(+), 2 deletions(-)

diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc
index 38d71111c9..17fcc21b0e 100644
--- a/tcg/riscv/tcg-target.c.inc
+++ b/tcg/riscv/tcg-target.c.inc
@@ -309,6 +309,12 @@ typedef enum {
     OPC_VS2R_V = 0x2000027 | V_UNIT_STRIDE_WHOLE_REG | V_NF(1),
     OPC_VS4R_V = 0x2000027 | V_UNIT_STRIDE_WHOLE_REG | V_NF(3),
     OPC_VS8R_V = 0x2000027 | V_UNIT_STRIDE_WHOLE_REG | V_NF(7),
+
+    OPC_VMV_V_V = 0x5e000057 | V_OPIVV,
+    OPC_VMV_V_I = 0x5e000057 | V_OPIVI,
+    OPC_VMV_V_X = 0x5e000057 | V_OPIVX,
+
+    OPC_VMVNR_V = 0x9e000057 | V_OPIVI,
 } RISCVInsn;
 
 /*
@@ -401,6 +407,16 @@ static int32_t encode_uj(RISCVInsn opc, TCGReg rd, uint32_t imm)
     return opc | (rd & 0x1f) << 7 | encode_ujimm20(imm);
 }
 
+
+/* Type-OPIVI */
+
+static int32_t encode_vi(RISCVInsn opc, TCGReg rd, int32_t imm,
+                         TCGReg vs2, bool vm)
+{
+    return opc | (rd & 0x1f) << 7 | (imm & 0x1f) << 15 |
+           (vs2 & 0x1f) << 20 | (vm << 25);
+}
+
 /* Type-OPIVV/OPMVV/OPIVX/OPMVX, Vector load and store */
 
 static int32_t encode_v(RISCVInsn opc, TCGReg d, TCGReg s1,
@@ -546,6 +562,24 @@ static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
  * RISC-V vector instruction emitters
  */
 
+/*
+ * Vector registers uses the same 5 lower bits as GPR registers,
+ * and vm=0 (vm = false) means vector masking ENABLED.
+ * With RVV 1.0, vs2 is the first operand, while rs1/imm is the
+ * second operand.
+ */
+static void tcg_out_opc_vx(TCGContext *s, RISCVInsn opc,
+                           TCGReg vd, TCGReg vs2, TCGReg rs1)
+{
+    tcg_out32(s, encode_v(opc, vd, rs1, vs2, true));
+}
+
+static void tcg_out_opc_vi(TCGContext *s, RISCVInsn opc,
+                           TCGReg vd, TCGReg vs2, int32_t imm)
+{
+    tcg_out32(s, encode_vi(opc, vd, imm, vs2, true));
+}
+
 typedef struct VsetCache {
     uint32_t movi_insn;
     uint32_t vset_insn;
@@ -574,6 +608,13 @@ static MemOp set_vtype_len(TCGContext *s, TCGType type)
     return s->riscv_cur_vsew;
 }
 
+static void set_vtype_len_sew(TCGContext *s, TCGType type, MemOp vsew)
+{
+    if (type != s->riscv_cur_type || vsew != s->riscv_cur_vsew) {
+        set_vtype(s, type, vsew);
+    }
+}
+
 /*
  * TCG intrinsics
  */
@@ -588,6 +629,15 @@ static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
     case TCG_TYPE_I64:
         tcg_out_opc_imm(s, OPC_ADDI, ret, arg, 0);
         break;
+    case TCG_TYPE_V64:
+    case TCG_TYPE_V128:
+    case TCG_TYPE_V256:
+        {
+            int lmul = type - riscv_lg2_vlenb;
+            int nf = 1 << MAX(lmul, 0);
+            tcg_out_opc_vi(s, OPC_VMVNR_V, ret, arg, nf - 1);
+        }
+        break;
     default:
         g_assert_not_reached();
     }
@@ -951,18 +1001,35 @@ static void tcg_out_addsub2(TCGContext *s,
 static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
                                    TCGReg dst, TCGReg src)
 {
-    return false;
+    set_vtype_len_sew(s, type, vece);
+    tcg_out_opc_vx(s, OPC_VMV_V_X, dst, 0, src);
+    return true;
 }
 
 static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
                                     TCGReg dst, TCGReg base, intptr_t offset)
 {
-    return false;
+    tcg_out_ld(s, TCG_TYPE_REG, TCG_REG_TMP0, base, offset);
+    return tcg_out_dup_vec(s, type, vece, dst, TCG_REG_TMP0);
 }
 
 static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
                                     TCGReg dst, int64_t arg)
 {
+    /* Arg is replicated by VECE; extract the highest element. */
+    arg >>= (-8 << vece) & 63;
+
+    if (arg >= -16 && arg < 16) {
+        if (arg == 0 || arg == -1) {
+            set_vtype_len(s, type);
+        } else {
+            set_vtype_len_sew(s, type, vece);
+        }
+        tcg_out_opc_vi(s, OPC_VMV_V_I, dst, 0, arg);
+        return;
+    }
+    tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, arg);
+    tcg_out_dup_vec(s, type, vece, dst, TCG_REG_TMP0);
 }
 
 static const struct {
@@ -2104,6 +2171,9 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
     a2 = args[2];
 
     switch (opc) {
+    case INDEX_op_dupm_vec:
+        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
+        break;
     case INDEX_op_ld_vec:
         tcg_out_ld(s, type, a0, a1, a2);
         break;
@@ -2272,6 +2342,8 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
 
     case INDEX_op_st_vec:
         return C_O0_I2(v, r);
+    case INDEX_op_dup_vec:
+    case INDEX_op_dupm_vec:
     case INDEX_op_ld_vec:
         return C_O1_I1(v, r);
     default:
-- 
2.43.0