[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[RFC PATCH v3 14/44] target/loongarch: Implement vmul/vmuh/vmulw{ev/od}
|
From: |
Song Gao |
|
Subject: |
[RFC PATCH v3 14/44] target/loongarch: Implement vmul/vmuh/vmulw{ev/od} |
|
Date: |
Thu, 20 Apr 2023 16:06:39 +0800 |
This patch includes:
- VMUL.{B/H/W/D};
- VMUH.{B/H/W/D}[U];
- VMULW{EV/OD}.{H.B/W.H/D.W/Q.D}[U];
- VMULW{EV/OD}.{H.BU.B/W.HU.H/D.WU.W/Q.DU.D}.
Signed-off-by: Song Gao <gaosong@loongson.cn>
---
include/tcg/tcg-op.h | 4 +
target/loongarch/disas.c | 38 ++
target/loongarch/helper.h | 30 ++
target/loongarch/insn_trans/trans_lsx.c.inc | 544 ++++++++++++++++++++
target/loongarch/insns.decode | 38 ++
target/loongarch/lsx_helper.c | 76 +++
tcg/tcg-op.c | 43 ++
7 files changed, 773 insertions(+)
diff --git a/include/tcg/tcg-op.h b/include/tcg/tcg-op.h
index dff17c7072..d90af683b5 100644
--- a/include/tcg/tcg-op.h
+++ b/include/tcg/tcg-op.h
@@ -321,6 +321,7 @@ void tcg_gen_sub2_i32(TCGv_i32 rl, TCGv_i32 rh, TCGv_i32 al,
void tcg_gen_mulu2_i32(TCGv_i32 rl, TCGv_i32 rh, TCGv_i32 arg1, TCGv_i32 arg2);
void tcg_gen_muls2_i32(TCGv_i32 rl, TCGv_i32 rh, TCGv_i32 arg1, TCGv_i32 arg2);
void tcg_gen_mulsu2_i32(TCGv_i32 rl, TCGv_i32 rh, TCGv_i32 arg1, TCGv_i32
arg2);
+void tcg_gen_mulus2_i32(TCGv_i32 rl, TCGv_i32 rh, TCGv_i32 arg1, TCGv_i32
arg2);
void tcg_gen_ext8s_i32(TCGv_i32 ret, TCGv_i32 arg);
void tcg_gen_ext16s_i32(TCGv_i32 ret, TCGv_i32 arg);
void tcg_gen_ext8u_i32(TCGv_i32 ret, TCGv_i32 arg);
@@ -517,6 +518,7 @@ void tcg_gen_sub2_i64(TCGv_i64 rl, TCGv_i64 rh, TCGv_i64 al,
void tcg_gen_mulu2_i64(TCGv_i64 rl, TCGv_i64 rh, TCGv_i64 arg1, TCGv_i64 arg2);
void tcg_gen_muls2_i64(TCGv_i64 rl, TCGv_i64 rh, TCGv_i64 arg1, TCGv_i64 arg2);
void tcg_gen_mulsu2_i64(TCGv_i64 rl, TCGv_i64 rh, TCGv_i64 arg1, TCGv_i64
arg2);
+void tcg_gen_mulus2_i64(TCGv_i64 rl, TCGv_i64 rh, TCGv_i64 arg1, TCGv_i64
arg2);
void tcg_gen_not_i64(TCGv_i64 ret, TCGv_i64 arg);
void tcg_gen_ext8s_i64(TCGv_i64 ret, TCGv_i64 arg);
void tcg_gen_ext16s_i64(TCGv_i64 ret, TCGv_i64 arg);
@@ -1096,6 +1098,7 @@ void tcg_gen_stl_vec(TCGv_vec r, TCGv_ptr base, TCGArg
offset, TCGType t);
#define tcg_gen_mulu2_tl tcg_gen_mulu2_i64
#define tcg_gen_muls2_tl tcg_gen_muls2_i64
#define tcg_gen_mulsu2_tl tcg_gen_mulsu2_i64
+#define tcg_gen_mulus2_tl tcg_gen_mulus2_i64
#define tcg_gen_smin_tl tcg_gen_smin_i64
#define tcg_gen_umin_tl tcg_gen_umin_i64
#define tcg_gen_smax_tl tcg_gen_smax_i64
@@ -1210,6 +1213,7 @@ void tcg_gen_stl_vec(TCGv_vec r, TCGv_ptr base, TCGArg
offset, TCGType t);
#define tcg_gen_mulu2_tl tcg_gen_mulu2_i32
#define tcg_gen_muls2_tl tcg_gen_muls2_i32
#define tcg_gen_mulsu2_tl tcg_gen_mulsu2_i32
+#define tcg-gen_mulus2_tl tcg_gen_mulus2_i32
#define tcg_gen_smin_tl tcg_gen_smin_i32
#define tcg_gen_umin_tl tcg_gen_umin_i32
#define tcg_gen_smax_tl tcg_gen_smax_i32
diff --git a/target/loongarch/disas.c b/target/loongarch/disas.c
index 6b0e518bfa..48e6ef5309 100644
--- a/target/loongarch/disas.c
+++ b/target/loongarch/disas.c
@@ -972,3 +972,41 @@ INSN_LSX(vmini_bu, vv_i)
INSN_LSX(vmini_hu, vv_i)
INSN_LSX(vmini_wu, vv_i)
INSN_LSX(vmini_du, vv_i)
+
+INSN_LSX(vmul_b, vvv)
+INSN_LSX(vmul_h, vvv)
+INSN_LSX(vmul_w, vvv)
+INSN_LSX(vmul_d, vvv)
+INSN_LSX(vmuh_b, vvv)
+INSN_LSX(vmuh_h, vvv)
+INSN_LSX(vmuh_w, vvv)
+INSN_LSX(vmuh_d, vvv)
+INSN_LSX(vmuh_bu, vvv)
+INSN_LSX(vmuh_hu, vvv)
+INSN_LSX(vmuh_wu, vvv)
+INSN_LSX(vmuh_du, vvv)
+
+INSN_LSX(vmulwev_h_b, vvv)
+INSN_LSX(vmulwev_w_h, vvv)
+INSN_LSX(vmulwev_d_w, vvv)
+INSN_LSX(vmulwev_q_d, vvv)
+INSN_LSX(vmulwod_h_b, vvv)
+INSN_LSX(vmulwod_w_h, vvv)
+INSN_LSX(vmulwod_d_w, vvv)
+INSN_LSX(vmulwod_q_d, vvv)
+INSN_LSX(vmulwev_h_bu, vvv)
+INSN_LSX(vmulwev_w_hu, vvv)
+INSN_LSX(vmulwev_d_wu, vvv)
+INSN_LSX(vmulwev_q_du, vvv)
+INSN_LSX(vmulwod_h_bu, vvv)
+INSN_LSX(vmulwod_w_hu, vvv)
+INSN_LSX(vmulwod_d_wu, vvv)
+INSN_LSX(vmulwod_q_du, vvv)
+INSN_LSX(vmulwev_h_bu_b, vvv)
+INSN_LSX(vmulwev_w_hu_h, vvv)
+INSN_LSX(vmulwev_d_wu_w, vvv)
+INSN_LSX(vmulwev_q_du_d, vvv)
+INSN_LSX(vmulwod_h_bu_b, vvv)
+INSN_LSX(vmulwod_w_hu_h, vvv)
+INSN_LSX(vmulwod_d_wu_w, vvv)
+INSN_LSX(vmulwod_q_du_d, vvv)
diff --git a/target/loongarch/helper.h b/target/loongarch/helper.h
index 379f961d1e..830d8cbe62 100644
--- a/target/loongarch/helper.h
+++ b/target/loongarch/helper.h
@@ -243,3 +243,33 @@ DEF_HELPER_FLAGS_4(vmaxi_bu, TCG_CALL_NO_RWG, void, ptr,
ptr, i64, i32)
DEF_HELPER_FLAGS_4(vmaxi_hu, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
DEF_HELPER_FLAGS_4(vmaxi_wu, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
DEF_HELPER_FLAGS_4(vmaxi_du, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
+
+DEF_HELPER_FLAGS_4(vmuh_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmuh_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmuh_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmuh_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmuh_bu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmuh_hu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmuh_wu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmuh_du, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(vmulwev_h_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwev_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwev_d_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_h_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_w_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_d_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(vmulwev_h_bu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwev_w_hu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwev_d_wu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_h_bu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_w_hu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_d_wu, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(vmulwev_h_bu_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwev_w_hu_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwev_d_wu_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_h_bu_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_w_hu_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(vmulwod_d_wu_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
diff --git a/target/loongarch/insn_trans/trans_lsx.c.inc
b/target/loongarch/insn_trans/trans_lsx.c.inc
index 15bb7888d3..81c1cd6291 100644
--- a/target/loongarch/insn_trans/trans_lsx.c.inc
+++ b/target/loongarch/insn_trans/trans_lsx.c.inc
@@ -1514,3 +1514,547 @@ TRANS(vmaxi_bu, gvec_vv_i, MO_8, do_vmaxi_u)
TRANS(vmaxi_hu, gvec_vv_i, MO_16, do_vmaxi_u)
TRANS(vmaxi_wu, gvec_vv_i, MO_32, do_vmaxi_u)
TRANS(vmaxi_du, gvec_vv_i, MO_64, do_vmaxi_u)
+
+TRANS(vmul_b, gvec_vvv, MO_8, tcg_gen_gvec_mul)
+TRANS(vmul_h, gvec_vvv, MO_16, tcg_gen_gvec_mul)
+TRANS(vmul_w, gvec_vvv, MO_32, tcg_gen_gvec_mul)
+TRANS(vmul_d, gvec_vvv, MO_64, tcg_gen_gvec_mul)
+
+static void gen_vmuh_w(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 discard = tcg_temp_new_i32();
+ tcg_gen_muls2_i32(discard, t, a, b);
+}
+
+static void gen_vmuh_d(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 discard = tcg_temp_new_i64();
+ tcg_gen_muls2_i64(discard, t, a, b);
+}
+
+static void do_vmuh_s(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const GVecGen3 op[4] = {
+ {
+ .fno = gen_helper_vmuh_b,
+ .vece = MO_8
+ },
+ {
+ .fno = gen_helper_vmuh_h,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmuh_w,
+ .fno = gen_helper_vmuh_w,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmuh_d,
+ .fno = gen_helper_vmuh_d,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmuh_b, gvec_vvv, MO_8, do_vmuh_s)
+TRANS(vmuh_h, gvec_vvv, MO_16, do_vmuh_s)
+TRANS(vmuh_w, gvec_vvv, MO_32, do_vmuh_s)
+TRANS(vmuh_d, gvec_vvv, MO_64, do_vmuh_s)
+
+static void gen_vmuh_wu(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 discard = tcg_temp_new_i32();
+ tcg_gen_mulu2_i32(discard, t, a, b);
+}
+
+static void gen_vmuh_du(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 discard = tcg_temp_new_i64();
+ tcg_gen_mulu2_i64(discard, t, a, b);
+}
+
+static void do_vmuh_u(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const GVecGen3 op[4] = {
+ {
+ .fno = gen_helper_vmuh_bu,
+ .vece = MO_8
+ },
+ {
+ .fno = gen_helper_vmuh_hu,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmuh_wu,
+ .fno = gen_helper_vmuh_wu,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmuh_du,
+ .fno = gen_helper_vmuh_du,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmuh_bu, gvec_vvv, MO_8, do_vmuh_u)
+TRANS(vmuh_hu, gvec_vvv, MO_16, do_vmuh_u)
+TRANS(vmuh_wu, gvec_vvv, MO_32, do_vmuh_u)
+TRANS(vmuh_du, gvec_vvv, MO_64, do_vmuh_u)
+
+static void gen_vmulwev_s(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t1, t2;
+ int halfbits = 4 << vece;
+
+ t1 = tcg_temp_new_vec_matching(a);
+ t2 = tcg_temp_new_vec_matching(b);
+ tcg_gen_shli_vec(vece, t1, a, halfbits);
+ tcg_gen_sari_vec(vece, t1, t1, halfbits);
+ tcg_gen_shli_vec(vece, t2, b, halfbits);
+ tcg_gen_sari_vec(vece, t2, t2, halfbits);
+ tcg_gen_mul_vec(vece, t, t1, t2);
+}
+
+static void gen_vmulwev_w_h(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t1, t2;
+
+ t1 = tcg_temp_new_i32();
+ t2 = tcg_temp_new_i32();
+ tcg_gen_ext16s_i32(t1, a);
+ tcg_gen_ext16s_i32(t2, b);
+ tcg_gen_mul_i32(t, t1, t2);
+}
+
+static void gen_vmulwev_d_w(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t1, t2;
+
+ t1 = tcg_temp_new_i64();
+ t2 = tcg_temp_new_i64();
+ tcg_gen_ext32s_i64(t1, a);
+ tcg_gen_ext32s_i64(t2, b);
+ tcg_gen_mul_i64(t, t1, t2);
+}
+
+static void do_vmulwev_s(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shli_vec, INDEX_op_sari_vec, INDEX_op_mul_vec, 0
+ };
+ static const GVecGen3 op[3] = {
+ {
+ .fniv = gen_vmulwev_s,
+ .fno = gen_helper_vmulwev_h_b,
+ .opt_opc = vecop_list,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmulwev_w_h,
+ .fniv = gen_vmulwev_s,
+ .fno = gen_helper_vmulwev_w_h,
+ .opt_opc = vecop_list,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmulwev_d_w,
+ .fniv = gen_vmulwev_s,
+ .fno = gen_helper_vmulwev_d_w,
+ .opt_opc = vecop_list,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmulwev_h_b, gvec_vvv, MO_8, do_vmulwev_s)
+TRANS(vmulwev_w_h, gvec_vvv, MO_16, do_vmulwev_s)
+TRANS(vmulwev_d_w, gvec_vvv, MO_32, do_vmulwev_s)
+
+#define VMUL_Q(NAME, FN, idx1, idx2) \
+static bool trans_## NAME (DisasContext *ctx, arg_vvv *a) \
+{ \
+ TCGv_i64 rh, rl, arg1, arg2; \
+ \
+ rh = tcg_temp_new_i64(); \
+ rl = tcg_temp_new_i64(); \
+ arg1 = tcg_temp_new_i64(); \
+ arg2 = tcg_temp_new_i64(); \
+ \
+ get_vreg64(arg1, a->vj, idx1); \
+ get_vreg64(arg2, a->vk, idx2); \
+ \
+ tcg_gen_## FN ##_i64(rl, rh, arg1, arg2); \
+ \
+ set_vreg64(rh, a->vd, 1); \
+ set_vreg64(rl, a->vd, 0); \
+ \
+ return true; \
+}
+
+VMUL_Q(vmulwev_q_d, muls2, 0, 0)
+VMUL_Q(vmulwod_q_d, muls2, 1, 1)
+VMUL_Q(vmulwev_q_du, mulu2, 0, 0)
+VMUL_Q(vmulwod_q_du, mulu2, 1, 1)
+VMUL_Q(vmulwev_q_du_d, mulus2, 0, 0)
+VMUL_Q(vmulwod_q_du_d, mulus2, 1, 1)
+
+static void gen_vmulwod_s(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t1, t2;
+ int halfbits = 4 << vece;
+
+ t1 = tcg_temp_new_vec_matching(a);
+ t2 = tcg_temp_new_vec_matching(b);
+ tcg_gen_sari_vec(vece, t1, a, halfbits);
+ tcg_gen_sari_vec(vece, t2, b, halfbits);
+ tcg_gen_mul_vec(vece, t, t1, t2);
+}
+
+static void gen_vmulwod_w_h(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t1, t2;
+
+ t1 = tcg_temp_new_i32();
+ t2 = tcg_temp_new_i32();
+ tcg_gen_sari_i32(t1, a, 16);
+ tcg_gen_sari_i32(t2, b, 16);
+ tcg_gen_mul_i32(t, t1, t2);
+}
+
+static void gen_vmulwod_d_w(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t1, t2;
+
+ t1 = tcg_temp_new_i64();
+ t2 = tcg_temp_new_i64();
+ tcg_gen_sari_i64(t1, a, 32);
+ tcg_gen_sari_i64(t2, b, 32);
+ tcg_gen_mul_i64(t, t1, t2);
+}
+
+static void do_vmulwod_s(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_sari_vec, INDEX_op_mul_vec, 0
+ };
+ static const GVecGen3 op[3] = {
+ {
+ .fniv = gen_vmulwod_s,
+ .fno = gen_helper_vmulwod_h_b,
+ .opt_opc = vecop_list,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmulwod_w_h,
+ .fniv = gen_vmulwod_s,
+ .fno = gen_helper_vmulwod_w_h,
+ .opt_opc = vecop_list,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmulwod_d_w,
+ .fniv = gen_vmulwod_s,
+ .fno = gen_helper_vmulwod_d_w,
+ .opt_opc = vecop_list,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmulwod_h_b, gvec_vvv, MO_8, do_vmulwod_s)
+TRANS(vmulwod_w_h, gvec_vvv, MO_16, do_vmulwod_s)
+TRANS(vmulwod_d_w, gvec_vvv, MO_32, do_vmulwod_s)
+
+static void gen_vmulwev_u(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t1, t2, mask;
+
+ t1 = tcg_temp_new_vec_matching(a);
+ t2 = tcg_temp_new_vec_matching(b);
+ mask = tcg_constant_vec_matching(t, vece, MAKE_64BIT_MASK(0, 4 << vece));
+ tcg_gen_and_vec(vece, t1, a, mask);
+ tcg_gen_and_vec(vece, t2, b, mask);
+ tcg_gen_mul_vec(vece, t, t1, t2);
+}
+
+static void gen_vmulwev_w_hu(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t1, t2;
+
+ t1 = tcg_temp_new_i32();
+ t2 = tcg_temp_new_i32();
+ tcg_gen_ext16u_i32(t1, a);
+ tcg_gen_ext16u_i32(t2, b);
+ tcg_gen_mul_i32(t, t1, t2);
+}
+
+static void gen_vmulwev_d_wu(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t1, t2;
+
+ t1 = tcg_temp_new_i64();
+ t2 = tcg_temp_new_i64();
+ tcg_gen_ext32u_i64(t1, a);
+ tcg_gen_ext32u_i64(t2, b);
+ tcg_gen_mul_i64(t, t1, t2);
+}
+
+static void do_vmulwev_u(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_mul_vec, 0
+ };
+ static const GVecGen3 op[3] = {
+ {
+ .fniv = gen_vmulwev_u,
+ .fno = gen_helper_vmulwev_h_bu,
+ .opt_opc = vecop_list,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmulwev_w_hu,
+ .fniv = gen_vmulwev_u,
+ .fno = gen_helper_vmulwev_w_hu,
+ .opt_opc = vecop_list,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmulwev_d_wu,
+ .fniv = gen_vmulwev_u,
+ .fno = gen_helper_vmulwev_d_wu,
+ .opt_opc = vecop_list,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmulwev_h_bu, gvec_vvv, MO_8, do_vmulwev_u)
+TRANS(vmulwev_w_hu, gvec_vvv, MO_16, do_vmulwev_u)
+TRANS(vmulwev_d_wu, gvec_vvv, MO_32, do_vmulwev_u)
+
+static void gen_vmulwod_u(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t1, t2;
+ int halfbits = 4 << vece;
+
+ t1 = tcg_temp_new_vec_matching(a);
+ t2 = tcg_temp_new_vec_matching(b);
+ tcg_gen_shri_vec(vece, t1, a, halfbits);
+ tcg_gen_shri_vec(vece, t2, b, halfbits);
+ tcg_gen_mul_vec(vece, t, t1, t2);
+}
+
+static void gen_vmulwod_w_hu(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t1, t2;
+
+ t1 = tcg_temp_new_i32();
+ t2 = tcg_temp_new_i32();
+ tcg_gen_shri_i32(t1, a, 16);
+ tcg_gen_shri_i32(t2, b, 16);
+ tcg_gen_mul_i32(t, t1, t2);
+}
+
+static void gen_vmulwod_d_wu(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t1, t2;
+
+ t1 = tcg_temp_new_i64();
+ t2 = tcg_temp_new_i64();
+ tcg_gen_shri_i64(t1, a, 32);
+ tcg_gen_shri_i64(t2, b, 32);
+ tcg_gen_mul_i64(t, t1, t2);
+}
+
+static void do_vmulwod_u(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_mul_vec, 0
+ };
+ static const GVecGen3 op[3] = {
+ {
+ .fniv = gen_vmulwod_u,
+ .fno = gen_helper_vmulwod_h_bu,
+ .opt_opc = vecop_list,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmulwod_w_hu,
+ .fniv = gen_vmulwod_u,
+ .fno = gen_helper_vmulwod_w_hu,
+ .opt_opc = vecop_list,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmulwod_d_wu,
+ .fniv = gen_vmulwod_u,
+ .fno = gen_helper_vmulwod_d_wu,
+ .opt_opc = vecop_list,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmulwod_h_bu, gvec_vvv, MO_8, do_vmulwod_u)
+TRANS(vmulwod_w_hu, gvec_vvv, MO_16, do_vmulwod_u)
+TRANS(vmulwod_d_wu, gvec_vvv, MO_32, do_vmulwod_u)
+
+static void gen_vmulwev_u_s(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t1, t2, mask;
+ int halfbits = 4 << vece;
+
+ t1 = tcg_temp_new_vec_matching(a);
+ t2 = tcg_temp_new_vec_matching(b);
+ mask = tcg_constant_vec_matching(t, vece, MAKE_64BIT_MASK(0, 4 << vece));
+ tcg_gen_and_vec(vece, t1, a, mask);
+ tcg_gen_shli_vec(vece, t2, b, halfbits);
+ tcg_gen_sari_vec(vece, t2, t2, halfbits);
+ tcg_gen_mul_vec(vece, t, t1, t2);
+}
+
+static void gen_vmulwev_w_hu_h(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t1, t2;
+
+ t1 = tcg_temp_new_i32();
+ t2 = tcg_temp_new_i32();
+ tcg_gen_ext16u_i32(t1, a);
+ tcg_gen_ext16s_i32(t2, b);
+ tcg_gen_mul_i32(t, t1, t2);
+}
+
+static void gen_vmulwev_d_wu_w(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t1, t2;
+
+ t1 = tcg_temp_new_i64();
+ t2 = tcg_temp_new_i64();
+ tcg_gen_ext32u_i64(t1, a);
+ tcg_gen_ext32s_i64(t2, b);
+ tcg_gen_mul_i64(t, t1, t2);
+}
+
+static void do_vmulwev_u_s(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shli_vec, INDEX_op_sari_vec, INDEX_op_mul_vec, 0
+ };
+ static const GVecGen3 op[3] = {
+ {
+ .fniv = gen_vmulwev_u_s,
+ .fno = gen_helper_vmulwev_h_bu_b,
+ .opt_opc = vecop_list,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmulwev_w_hu_h,
+ .fniv = gen_vmulwev_u_s,
+ .fno = gen_helper_vmulwev_w_hu_h,
+ .opt_opc = vecop_list,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmulwev_d_wu_w,
+ .fniv = gen_vmulwev_u_s,
+ .fno = gen_helper_vmulwev_d_wu_w,
+ .opt_opc = vecop_list,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmulwev_h_bu_b, gvec_vvv, MO_8, do_vmulwev_u_s)
+TRANS(vmulwev_w_hu_h, gvec_vvv, MO_16, do_vmulwev_u_s)
+TRANS(vmulwev_d_wu_w, gvec_vvv, MO_32, do_vmulwev_u_s)
+
+static void gen_vmulwod_u_s(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+ TCGv_vec t1, t2;
+ int halfbits = 4 << vece;
+
+ t1 = tcg_temp_new_vec_matching(a);
+ t2 = tcg_temp_new_vec_matching(b);
+ tcg_gen_shri_vec(vece, t1, a, halfbits);
+ tcg_gen_sari_vec(vece, t2, b, halfbits);
+ tcg_gen_mul_vec(vece, t, t1, t2);
+}
+
+static void gen_vmulwod_w_hu_h(TCGv_i32 t, TCGv_i32 a, TCGv_i32 b)
+{
+ TCGv_i32 t1, t2;
+
+ t1 = tcg_temp_new_i32();
+ t2 = tcg_temp_new_i32();
+ tcg_gen_shri_i32(t1, a, 16);
+ tcg_gen_sari_i32(t2, b, 16);
+ tcg_gen_mul_i32(t, t1, t2);
+}
+static void gen_vmulwod_d_wu_w(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+ TCGv_i64 t1, t2;
+
+ t1 = tcg_temp_new_i64();
+ t2 = tcg_temp_new_i64();
+ tcg_gen_shri_i64(t1, a, 32);
+ tcg_gen_sari_i64(t2, b, 32);
+ tcg_gen_mul_i64(t, t1, t2);
+}
+
+static void do_vmulwod_u_s(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ uint32_t vk_ofs, uint32_t oprsz, uint32_t maxsz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_sari_vec, INDEX_op_mul_vec, 0
+ };
+ static const GVecGen3 op[3] = {
+ {
+ .fniv = gen_vmulwod_u_s,
+ .fno = gen_helper_vmulwod_h_bu_b,
+ .opt_opc = vecop_list,
+ .vece = MO_16
+ },
+ {
+ .fni4 = gen_vmulwod_w_hu_h,
+ .fniv = gen_vmulwod_u_s,
+ .fno = gen_helper_vmulwod_w_hu_h,
+ .opt_opc = vecop_list,
+ .vece = MO_32
+ },
+ {
+ .fni8 = gen_vmulwod_d_wu_w,
+ .fniv = gen_vmulwod_u_s,
+ .fno = gen_helper_vmulwod_d_wu_w,
+ .opt_opc = vecop_list,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_3(vd_ofs, vj_ofs, vk_ofs, oprsz, maxsz, &op[vece]);
+}
+
+TRANS(vmulwod_h_bu_b, gvec_vvv, MO_8, do_vmulwod_u_s)
+TRANS(vmulwod_w_hu_h, gvec_vvv, MO_16, do_vmulwod_u_s)
+TRANS(vmulwod_d_wu_w, gvec_vvv, MO_32, do_vmulwod_u_s)
diff --git a/target/loongarch/insns.decode b/target/loongarch/insns.decode
index dd1bc031e8..64e8042c9c 100644
--- a/target/loongarch/insns.decode
+++ b/target/loongarch/insns.decode
@@ -668,3 +668,41 @@ vmini_bu 0111 00101001 01100 ..... ..... .....
@vv_ui5
vmini_hu 0111 00101001 01101 ..... ..... ..... @vv_ui5
vmini_wu 0111 00101001 01110 ..... ..... ..... @vv_ui5
vmini_du 0111 00101001 01111 ..... ..... ..... @vv_ui5
+
+vmul_b 0111 00001000 01000 ..... ..... ..... @vvv
+vmul_h 0111 00001000 01001 ..... ..... ..... @vvv
+vmul_w 0111 00001000 01010 ..... ..... ..... @vvv
+vmul_d 0111 00001000 01011 ..... ..... ..... @vvv
+vmuh_b 0111 00001000 01100 ..... ..... ..... @vvv
+vmuh_h 0111 00001000 01101 ..... ..... ..... @vvv
+vmuh_w 0111 00001000 01110 ..... ..... ..... @vvv
+vmuh_d 0111 00001000 01111 ..... ..... ..... @vvv
+vmuh_bu 0111 00001000 10000 ..... ..... ..... @vvv
+vmuh_hu 0111 00001000 10001 ..... ..... ..... @vvv
+vmuh_wu 0111 00001000 10010 ..... ..... ..... @vvv
+vmuh_du 0111 00001000 10011 ..... ..... ..... @vvv
+
+vmulwev_h_b 0111 00001001 00000 ..... ..... ..... @vvv
+vmulwev_w_h 0111 00001001 00001 ..... ..... ..... @vvv
+vmulwev_d_w 0111 00001001 00010 ..... ..... ..... @vvv
+vmulwev_q_d 0111 00001001 00011 ..... ..... ..... @vvv
+vmulwod_h_b 0111 00001001 00100 ..... ..... ..... @vvv
+vmulwod_w_h 0111 00001001 00101 ..... ..... ..... @vvv
+vmulwod_d_w 0111 00001001 00110 ..... ..... ..... @vvv
+vmulwod_q_d 0111 00001001 00111 ..... ..... ..... @vvv
+vmulwev_h_bu 0111 00001001 10000 ..... ..... ..... @vvv
+vmulwev_w_hu 0111 00001001 10001 ..... ..... ..... @vvv
+vmulwev_d_wu 0111 00001001 10010 ..... ..... ..... @vvv
+vmulwev_q_du 0111 00001001 10011 ..... ..... ..... @vvv
+vmulwod_h_bu 0111 00001001 10100 ..... ..... ..... @vvv
+vmulwod_w_hu 0111 00001001 10101 ..... ..... ..... @vvv
+vmulwod_d_wu 0111 00001001 10110 ..... ..... ..... @vvv
+vmulwod_q_du 0111 00001001 10111 ..... ..... ..... @vvv
+vmulwev_h_bu_b 0111 00001010 00000 ..... ..... ..... @vvv
+vmulwev_w_hu_h 0111 00001010 00001 ..... ..... ..... @vvv
+vmulwev_d_wu_w 0111 00001010 00010 ..... ..... ..... @vvv
+vmulwev_q_du_d 0111 00001010 00011 ..... ..... ..... @vvv
+vmulwod_h_bu_b 0111 00001010 00100 ..... ..... ..... @vvv
+vmulwod_w_hu_h 0111 00001010 00101 ..... ..... ..... @vvv
+vmulwod_d_wu_w 0111 00001010 00110 ..... ..... ..... @vvv
+vmulwod_q_du_d 0111 00001010 00111 ..... ..... ..... @vvv
diff --git a/target/loongarch/lsx_helper.c b/target/loongarch/lsx_helper.c
index 40f52bf07d..3b565277fd 100644
--- a/target/loongarch/lsx_helper.c
+++ b/target/loongarch/lsx_helper.c
@@ -370,3 +370,79 @@ VMINMAXI(vmaxi_bu, 8, UB, DO_MAX)
VMINMAXI(vmaxi_hu, 16, UH, DO_MAX)
VMINMAXI(vmaxi_wu, 32, UW, DO_MAX)
VMINMAXI(vmaxi_du, 64, UD, DO_MAX)
+
+#define DO_VMUH(NAME, BIT, E1, E2, DO_OP) \
+void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t v) \
+{ \
+ int i; \
+ VReg *Vd = (VReg *)vd; \
+ VReg *Vj = (VReg *)vj; \
+ VReg *Vk = (VReg *)vk; \
+ typedef __typeof(Vd->E1(0)) T; \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ Vd->E2(i) = ((T)Vj->E2(i)) * ((T)Vk->E2(i)) >> BIT; \
+ } \
+}
+
+void HELPER(vmuh_d)(void *vd, void *vj, void *vk, uint32_t v)
+{
+ uint64_t l, h1, h2;
+ VReg *Vd = (VReg *)vd;
+ VReg *Vj = (VReg *)vj;
+ VReg *Vk = (VReg *)vk;
+
+ muls64(&l, &h1, Vj->D(0), Vk->D(0));
+ muls64(&l, &h2, Vj->D(1), Vk->D(1));
+
+ Vd->D(0) = h1;
+ Vd->D(1) = h2;
+}
+
+DO_VMUH(vmuh_b, 8, H, B, DO_MUH)
+DO_VMUH(vmuh_h, 16, W, H, DO_MUH)
+DO_VMUH(vmuh_w, 32, D, W, DO_MUH)
+
+void HELPER(vmuh_du)(void *vd, void *vj, void *vk, uint32_t v)
+{
+ uint64_t l, h1, h2;
+ VReg *Vd = (VReg *)vd;
+ VReg *Vj = (VReg *)vj;
+ VReg *Vk = (VReg *)vk;
+
+ mulu64(&l, &h1, Vj->D(0), Vk->D(0));
+ mulu64(&l, &h2, Vj->D(1), Vk->D(1));
+
+ Vd->D(0) = h1;
+ Vd->D(1) = h2;
+}
+
+DO_VMUH(vmuh_bu, 8, UH, UB, DO_MUH)
+DO_VMUH(vmuh_hu, 16, UW, UH, DO_MUH)
+DO_VMUH(vmuh_wu, 32, UD, UW, DO_MUH)
+
+#define DO_MUL(a, b) (a * b)
+
+DO_EVEN(vmulwev_h_b, 16, H, B, DO_MUL)
+DO_EVEN(vmulwev_w_h, 32, W, H, DO_MUL)
+DO_EVEN(vmulwev_d_w, 64, D, W, DO_MUL)
+
+DO_ODD(vmulwod_h_b, 16, H, B, DO_MUL)
+DO_ODD(vmulwod_w_h, 32, W, H, DO_MUL)
+DO_ODD(vmulwod_d_w, 64, D, W, DO_MUL)
+
+DO_EVEN(vmulwev_h_bu, 16, UH, UB, DO_MUL)
+DO_EVEN(vmulwev_w_hu, 32, UW, UH, DO_MUL)
+DO_EVEN(vmulwev_d_wu, 64, UD, UW, DO_MUL)
+
+DO_ODD(vmulwod_h_bu, 16, UH, UB, DO_MUL)
+DO_ODD(vmulwod_w_hu, 32, UW, UH, DO_MUL)
+DO_ODD(vmulwod_d_wu, 64, UD, UW, DO_MUL)
+
+DO_EVEN_U_S(vmulwev_h_bu_b, 16, H, UH, B, UB, DO_MUL)
+DO_EVEN_U_S(vmulwev_w_hu_h, 32, W, UW, H, UH, DO_MUL)
+DO_EVEN_U_S(vmulwev_d_wu_w, 64, D, UD, W, UW, DO_MUL)
+
+DO_ODD_U_S(vmulwod_h_bu_b, 16, H, UH, B, UB, DO_MUL)
+DO_ODD_U_S(vmulwod_w_hu_h, 32, W, UW, H, UH, DO_MUL)
+DO_ODD_U_S(vmulwod_d_wu_w, 64, D, UD, W, UW, DO_MUL)
diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c
index 3136cef81a..97abe6e99a 100644
--- a/tcg/tcg-op.c
+++ b/tcg/tcg-op.c
@@ -972,6 +972,33 @@ void tcg_gen_mulsu2_i32(TCGv_i32 rl, TCGv_i32 rh, TCGv_i32
arg1, TCGv_i32 arg2)
}
}
+void tcg_gen_mulus2_i32(TCGv_i32 rl, TCGv_i32 rh, TCGv_i32 arg1, TCGv_i32 arg2)
+{
+ if (TCG_TARGET_REG_BITS == 32) {
+ TCGv_i32 t0 = tcg_temp_ebb_new_i32();
+ TCGv_i32 t1 = tcg_temp_ebb_new_i32();
+ TCGv_i32 t2 = tcg_temp_ebb_new_i32();
+ tcg_gen_mulu2_i32(t0, t1, arg1, arg2);
+ /* Adjust for negative input for the signed arg2. */
+ tcg_gen_sari_i32(t2, arg2, 31);
+ tcg_gen_and_i32(t2, t2, arg1);
+ tcg_gen_sub_i32(rh, t1, t2);
+ tcg_gen_mov_i32(rl, t0);
+ tcg_temp_free_i32(t0);
+ tcg_temp_free_i32(t1);
+ tcg_temp_free_i32(t2);
+ } else {
+ TCGv_i64 t0 = tcg_temp_ebb_new_i64();
+ TCGv_i64 t1 = tcg_temp_ebb_new_i64();
+ tcg_gen_extu_i32_i64(t0, arg1);
+ tcg_gen_ext_i32_i64(t1, arg2);
+ tcg_gen_mul_i64(t0, t0, t1);
+ tcg_gen_extr_i64_i32(rl, rh, t0);
+ tcg_temp_free_i64(t0);
+ tcg_temp_free_i64(t1);
+ }
+}
+
void tcg_gen_ext8s_i32(TCGv_i32 ret, TCGv_i32 arg)
{
if (TCG_TARGET_HAS_ext8s_i32) {
@@ -2634,6 +2661,22 @@ void tcg_gen_mulsu2_i64(TCGv_i64 rl, TCGv_i64 rh,
TCGv_i64 arg1, TCGv_i64 arg2)
tcg_temp_free_i64(t2);
}
+void tcg_gen_mulus2_i64(TCGv_i64 rl, TCGv_i64 rh, TCGv_i64 arg1, TCGv_i64 arg2)
+{
+ TCGv_i64 t0 = tcg_temp_ebb_new_i64();
+ TCGv_i64 t1 = tcg_temp_ebb_new_i64();
+ TCGv_i64 t2 = tcg_temp_ebb_new_i64();
+ tcg_gen_mulu2_i64(t0, t1, arg1, arg2);
+ /* Adjust for negative input for the signed arg2. */
+ tcg_gen_sari_i64(t2, arg2, 63);
+ tcg_gen_and_i64(t2, t2, arg1);
+ tcg_gen_sub_i64(rh, t1, t2);
+ tcg_gen_mov_i64(rl, t0);
+ tcg_temp_free_i64(t0);
+ tcg_temp_free_i64(t1);
+ tcg_temp_free_i64(t2);
+}
+
void tcg_gen_smin_i64(TCGv_i64 ret, TCGv_i64 a, TCGv_i64 b)
{
tcg_gen_movcond_i64(TCG_COND_LT, ret, a, b, a, b);
--
2.31.1
- [RFC PATCH v3 04/44] target/loongarch: Implement vadd/vsub, (continued)
- [RFC PATCH v3 04/44] target/loongarch: Implement vadd/vsub, Song Gao, 2023/04/20
- [RFC PATCH v3 05/44] target/loongarch: Implement vaddi/vsubi, Song Gao, 2023/04/20
- [RFC PATCH v3 07/44] target/loongarch: Implement vsadd/vssub, Song Gao, 2023/04/20
- [RFC PATCH v3 08/44] target/loongarch: Implement vhaddw/vhsubw, Song Gao, 2023/04/20
- [RFC PATCH v3 01/44] target/loongarch: Add LSX data type VReg, Song Gao, 2023/04/20
- [RFC PATCH v3 09/44] target/loongarch: Implement vaddw/vsubw, Song Gao, 2023/04/20
- [RFC PATCH v3 11/44] target/loongarch: Implement vabsd, Song Gao, 2023/04/20
- [RFC PATCH v3 14/44] target/loongarch: Implement vmul/vmuh/vmulw{ev/od},
Song Gao <=
[RFC PATCH v3 16/44] target/loongarch: Implement vdiv/vmod, Song Gao, 2023/04/20
[RFC PATCH v3 13/44] target/loongarch: Implement vmax/vmin, Song Gao, 2023/04/20
[RFC PATCH v3 15/44] target/loongarch: Implement vmadd/vmsub/vmaddw{ev/od}, Song Gao, 2023/04/20
[RFC PATCH v3 20/44] target/loongarch: Implement vmskltz/vmskgez/vmsknz, Song Gao, 2023/04/20