target/loongarch: Implement xvreplve xvinsve0 xvpickve
This patch includes:
- XVREPLVE.{B/H/W/D};
- XVREPL128VEI.{B/H/W/D};
- XVREPLVE0.{B/H/W/D/Q};
- XVINSVE0.{W/D};
- XVPICKVE.{W/D};
- XVBSLL.V, XVBSRL.V.
Signed-off-by: Song Gao <gaosong@loongson.cn>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20230914022645.1151356-53-gaosong@loongson.cn>
diff --git a/target/loongarch/disas.c b/target/loongarch/disas.c
index 04f9f9f..d091402 100644
--- a/target/loongarch/disas.c
+++ b/target/loongarch/disas.c
@@ -1748,6 +1748,11 @@
output(ctx, mnemonic, "r%d, x%d, 0x%x", a->rd, a->vj, a->imm);
}
+static void output_vvr_x(DisasContext *ctx, arg_vvr *a, const char *mnemonic)
+{
+ output(ctx, mnemonic, "x%d, x%d, r%d", a->vd, a->vj, a->rk);
+}
+
INSN_LASX(xvadd_b, vvv)
INSN_LASX(xvadd_h, vvv)
INSN_LASX(xvadd_w, vvv)
@@ -2518,3 +2523,27 @@
INSN_LASX(xvreplgr2vr_h, vr)
INSN_LASX(xvreplgr2vr_w, vr)
INSN_LASX(xvreplgr2vr_d, vr)
+
+INSN_LASX(xvreplve_b, vvr)
+INSN_LASX(xvreplve_h, vvr)
+INSN_LASX(xvreplve_w, vvr)
+INSN_LASX(xvreplve_d, vvr)
+INSN_LASX(xvrepl128vei_b, vv_i)
+INSN_LASX(xvrepl128vei_h, vv_i)
+INSN_LASX(xvrepl128vei_w, vv_i)
+INSN_LASX(xvrepl128vei_d, vv_i)
+
+INSN_LASX(xvreplve0_b, vv)
+INSN_LASX(xvreplve0_h, vv)
+INSN_LASX(xvreplve0_w, vv)
+INSN_LASX(xvreplve0_d, vv)
+INSN_LASX(xvreplve0_q, vv)
+
+INSN_LASX(xvinsve0_w, vv_i)
+INSN_LASX(xvinsve0_d, vv_i)
+
+INSN_LASX(xvpickve_w, vv_i)
+INSN_LASX(xvpickve_d, vv_i)
+
+INSN_LASX(xvbsll_v, vv_i)
+INSN_LASX(xvbsrl_v, vv_i)
diff --git a/target/loongarch/helper.h b/target/loongarch/helper.h
index 8523358..fb489dd 100644
--- a/target/loongarch/helper.h
+++ b/target/loongarch/helper.h
@@ -668,6 +668,11 @@
DEF_HELPER_4(vsetallnez_w, void, env, i32, i32, i32)
DEF_HELPER_4(vsetallnez_d, void, env, i32, i32, i32)
+DEF_HELPER_FLAGS_4(xvinsve0_w, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
+DEF_HELPER_FLAGS_4(xvinsve0_d, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
+DEF_HELPER_FLAGS_4(xvpickve_w, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
+DEF_HELPER_FLAGS_4(xvpickve_d, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
+
DEF_HELPER_FLAGS_4(vpackev_b, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(vpackev_h, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(vpackev_w, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
diff --git a/target/loongarch/insn_trans/trans_vec.c.inc b/target/loongarch/insn_trans/trans_vec.c.inc
index e1ba540..ab3d92d 100644
--- a/target/loongarch/insn_trans/trans_vec.c.inc
+++ b/target/loongarch/insn_trans/trans_vec.c.inc
@@ -5000,112 +5000,169 @@
return true;
}
-static bool gen_vreplve(DisasContext *ctx, arg_vvr *a, int vece, int bit,
- void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long))
+static bool gen_vreplve_vl(DisasContext *ctx, arg_vvr *a,
+ uint32_t oprsz, int vece, int bit,
+ void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long))
{
+ int i;
TCGv_i64 t0 = tcg_temp_new_i64();
TCGv_ptr t1 = tcg_temp_new_ptr();
TCGv_i64 t2 = tcg_temp_new_i64();
- if (!avail_LSX(ctx)) {
- return false;
- }
-
- if (!check_vec(ctx, 16)) {
+ if (!check_vec(ctx, oprsz)) {
return true;
}
- tcg_gen_andi_i64(t0, gpr_src(ctx, a->rk, EXT_NONE), (LSX_LEN/bit) -1);
+ tcg_gen_andi_i64(t0, gpr_src(ctx, a->rk, EXT_NONE), (LSX_LEN / bit) - 1);
tcg_gen_shli_i64(t0, t0, vece);
if (HOST_BIG_ENDIAN) {
- tcg_gen_xori_i64(t0, t0, vece << ((LSX_LEN/bit) -1));
+ tcg_gen_xori_i64(t0, t0, vece << ((LSX_LEN / bit) - 1));
}
tcg_gen_trunc_i64_ptr(t1, t0);
tcg_gen_add_ptr(t1, t1, cpu_env);
- func(t2, t1, vec_full_offset(a->vj));
- tcg_gen_gvec_dup_i64(vece, vec_full_offset(a->vd), 16, ctx->vl/8, t2);
+
+ for (i = 0; i < oprsz; i += 16) {
+ func(t2, t1, vec_full_offset(a->vj) + i);
+ tcg_gen_gvec_dup_i64(vece, vec_full_offset(a->vd) + i, 16, 16, t2);
+ }
return true;
}
+static bool gen_vreplve(DisasContext *ctx, arg_vvr *a, int vece, int bit,
+ void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long))
+{
+ return gen_vreplve_vl(ctx, a, 16, vece, bit, func);
+}
+
+static bool gen_xvreplve(DisasContext *ctx, arg_vvr *a, int vece, int bit,
+ void (*func)(TCGv_i64, TCGv_ptr, tcg_target_long))
+{
+ return gen_vreplve_vl(ctx, a, 32, vece, bit, func);
+}
+
TRANS(vreplve_b, LSX, gen_vreplve, MO_8, 8, tcg_gen_ld8u_i64)
TRANS(vreplve_h, LSX, gen_vreplve, MO_16, 16, tcg_gen_ld16u_i64)
TRANS(vreplve_w, LSX, gen_vreplve, MO_32, 32, tcg_gen_ld32u_i64)
TRANS(vreplve_d, LSX, gen_vreplve, MO_64, 64, tcg_gen_ld_i64)
+TRANS(xvreplve_b, LASX, gen_xvreplve, MO_8, 8, tcg_gen_ld8u_i64)
+TRANS(xvreplve_h, LASX, gen_xvreplve, MO_16, 16, tcg_gen_ld16u_i64)
+TRANS(xvreplve_w, LASX, gen_xvreplve, MO_32, 32, tcg_gen_ld32u_i64)
+TRANS(xvreplve_d, LASX, gen_xvreplve, MO_64, 64, tcg_gen_ld_i64)
-static bool trans_vbsll_v(DisasContext *ctx, arg_vv_i *a)
+static bool gen_xvrepl128(DisasContext *ctx, arg_vv_i *a, MemOp mop)
{
- int ofs;
- TCGv_i64 desthigh, destlow, high, low;
+ int i;
- if (!avail_LSX(ctx)) {
- return false;
- }
-
- if (!check_vec(ctx, 16)) {
+ if (!check_vec(ctx, 32)) {
return true;
}
- desthigh = tcg_temp_new_i64();
- destlow = tcg_temp_new_i64();
- high = tcg_temp_new_i64();
- low = tcg_temp_new_i64();
+ for (i = 0; i < 32; i += 16) {
+ tcg_gen_gvec_dup_mem(mop, vec_full_offset(a->vd) + i,
+ vec_reg_offset(a->vj, a->imm, mop) + i, 16, 16);
- get_vreg64(low, a->vj, 0);
+ }
+ return true;
+}
- ofs = ((a->imm) & 0xf) * 8;
- if (ofs < 64) {
- get_vreg64(high, a->vj, 1);
- tcg_gen_extract2_i64(desthigh, low, high, 64 - ofs);
- tcg_gen_shli_i64(destlow, low, ofs);
- } else {
- tcg_gen_shli_i64(desthigh, low, ofs - 64);
- destlow = tcg_constant_i64(0);
+TRANS(xvrepl128vei_b, LASX, gen_xvrepl128, MO_8)
+TRANS(xvrepl128vei_h, LASX, gen_xvrepl128, MO_16)
+TRANS(xvrepl128vei_w, LASX, gen_xvrepl128, MO_32)
+TRANS(xvrepl128vei_d, LASX, gen_xvrepl128, MO_64)
+
+static bool gen_xvreplve0(DisasContext *ctx, arg_vv *a, MemOp mop)
+{
+ if (!check_vec(ctx, 32)) {
+ return true;
}
- set_vreg64(desthigh, a->vd, 1);
- set_vreg64(destlow, a->vd, 0);
+ tcg_gen_gvec_dup_mem(mop, vec_full_offset(a->vd),
+ vec_full_offset(a->vj), 32, 32);
+ return true;
+}
+
+TRANS(xvreplve0_b, LASX, gen_xvreplve0, MO_8)
+TRANS(xvreplve0_h, LASX, gen_xvreplve0, MO_16)
+TRANS(xvreplve0_w, LASX, gen_xvreplve0, MO_32)
+TRANS(xvreplve0_d, LASX, gen_xvreplve0, MO_64)
+TRANS(xvreplve0_q, LASX, gen_xvreplve0, MO_128)
+
+TRANS(xvinsve0_w, LASX, gen_xx_i, gen_helper_xvinsve0_w)
+TRANS(xvinsve0_d, LASX, gen_xx_i, gen_helper_xvinsve0_d)
+
+TRANS(xvpickve_w, LASX, gen_xx_i, gen_helper_xvpickve_w)
+TRANS(xvpickve_d, LASX, gen_xx_i, gen_helper_xvpickve_d)
+
+static bool do_vbsll_v(DisasContext *ctx, arg_vv_i *a, uint32_t oprsz)
+{
+ int i, ofs;
+
+ if (!check_vec(ctx, oprsz)) {
+ return true;
+ }
+
+ for (i = 0; i < oprsz / 16; i++) {
+ TCGv desthigh = tcg_temp_new_i64();
+ TCGv destlow = tcg_temp_new_i64();
+ TCGv high = tcg_temp_new_i64();
+ TCGv low = tcg_temp_new_i64();
+
+ get_vreg64(low, a->vj, 2 * i);
+
+ ofs = ((a->imm) & 0xf) * 8;
+ if (ofs < 64) {
+ get_vreg64(high, a->vj, 2 * i + 1);
+ tcg_gen_extract2_i64(desthigh, low, high, 64 - ofs);
+ tcg_gen_shli_i64(destlow, low, ofs);
+ } else {
+ tcg_gen_shli_i64(desthigh, low, ofs - 64);
+ destlow = tcg_constant_i64(0);
+ }
+ set_vreg64(desthigh, a->vd, 2 * i + 1);
+ set_vreg64(destlow, a->vd, 2 * i);
+ }
return true;
}
-static bool trans_vbsrl_v(DisasContext *ctx, arg_vv_i *a)
+static bool do_vbsrl_v(DisasContext *ctx, arg_vv_i *a, uint32_t oprsz)
{
- TCGv_i64 desthigh, destlow, high, low;
- int ofs;
+ int i, ofs;
- if (!avail_LSX(ctx)) {
- return false;
- }
-
- if (!check_vec(ctx, 16)) {
+ if (!check_vec(ctx, 32)) {
return true;
}
- desthigh = tcg_temp_new_i64();
- destlow = tcg_temp_new_i64();
- high = tcg_temp_new_i64();
- low = tcg_temp_new_i64();
+ for (i = 0; i < oprsz / 16; i++) {
+ TCGv desthigh = tcg_temp_new_i64();
+ TCGv destlow = tcg_temp_new_i64();
+ TCGv high = tcg_temp_new_i64();
+ TCGv low = tcg_temp_new_i64();
+ get_vreg64(high, a->vj, 2 * i + 1);
- get_vreg64(high, a->vj, 1);
-
- ofs = ((a->imm) & 0xf) * 8;
- if (ofs < 64) {
- get_vreg64(low, a->vj, 0);
- tcg_gen_extract2_i64(destlow, low, high, ofs);
- tcg_gen_shri_i64(desthigh, high, ofs);
- } else {
- tcg_gen_shri_i64(destlow, high, ofs - 64);
- desthigh = tcg_constant_i64(0);
+ ofs = ((a->imm) & 0xf) * 8;
+ if (ofs < 64) {
+ get_vreg64(low, a->vj, 2 * i);
+ tcg_gen_extract2_i64(destlow, low, high, ofs);
+ tcg_gen_shri_i64(desthigh, high, ofs);
+ } else {
+ tcg_gen_shri_i64(destlow, high, ofs - 64);
+ desthigh = tcg_constant_i64(0);
+ }
+ set_vreg64(desthigh, a->vd, 2 * i + 1);
+ set_vreg64(destlow, a->vd, 2 * i);
}
- set_vreg64(desthigh, a->vd, 1);
- set_vreg64(destlow, a->vd, 0);
-
return true;
}
+TRANS(vbsll_v, LSX, do_vbsll_v, 16)
+TRANS(vbsrl_v, LSX, do_vbsrl_v, 16)
+TRANS(xvbsll_v, LASX, do_vbsll_v, 32)
+TRANS(xvbsrl_v, LASX, do_vbsrl_v, 32)
+
TRANS(vpackev_b, LSX, gen_vvv, gen_helper_vpackev_b)
TRANS(vpackev_h, LSX, gen_vvv, gen_helper_vpackev_h)
TRANS(vpackev_w, LSX, gen_vvv, gen_helper_vpackev_w)
diff --git a/target/loongarch/insns.decode b/target/loongarch/insns.decode
index bb3bb44..74383ba 100644
--- a/target/loongarch/insns.decode
+++ b/target/loongarch/insns.decode
@@ -1987,3 +1987,28 @@
xvreplgr2vr_h 0111 01101001 11110 00001 ..... ..... @vr
xvreplgr2vr_w 0111 01101001 11110 00010 ..... ..... @vr
xvreplgr2vr_d 0111 01101001 11110 00011 ..... ..... @vr
+
+xvreplve_b 0111 01010010 00100 ..... ..... ..... @vvr
+xvreplve_h 0111 01010010 00101 ..... ..... ..... @vvr
+xvreplve_w 0111 01010010 00110 ..... ..... ..... @vvr
+xvreplve_d 0111 01010010 00111 ..... ..... ..... @vvr
+
+xvrepl128vei_b 0111 01101111 01111 0 .... ..... ..... @vv_ui4
+xvrepl128vei_h 0111 01101111 01111 10 ... ..... ..... @vv_ui3
+xvrepl128vei_w 0111 01101111 01111 110 .. ..... ..... @vv_ui2
+xvrepl128vei_d 0111 01101111 01111 1110 . ..... ..... @vv_ui1
+
+xvreplve0_b 0111 01110000 01110 00000 ..... ..... @vv
+xvreplve0_h 0111 01110000 01111 00000 ..... ..... @vv
+xvreplve0_w 0111 01110000 01111 10000 ..... ..... @vv
+xvreplve0_d 0111 01110000 01111 11000 ..... ..... @vv
+xvreplve0_q 0111 01110000 01111 11100 ..... ..... @vv
+
+xvinsve0_w 0111 01101111 11111 10 ... ..... ..... @vv_ui3
+xvinsve0_d 0111 01101111 11111 110 .. ..... ..... @vv_ui2
+
+xvpickve_w 0111 01110000 00111 10 ... ..... ..... @vv_ui3
+xvpickve_d 0111 01110000 00111 110 .. ..... ..... @vv_ui2
+
+xvbsll_v 0111 01101000 11100 ..... ..... ..... @vv_ui5
+xvbsrl_v 0111 01101000 11101 ..... ..... ..... @vv_ui5
diff --git a/target/loongarch/vec_helper.c b/target/loongarch/vec_helper.c
index 1179397..26d48ed 100644
--- a/target/loongarch/vec_helper.c
+++ b/target/loongarch/vec_helper.c
@@ -3209,6 +3209,34 @@
SETALLNEZ(vsetallnez_w, MO_32)
SETALLNEZ(vsetallnez_d, MO_64)
+#define XVINSVE0(NAME, E, MASK) \
+void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
+{ \
+ VReg *Vd = (VReg *)vd; \
+ VReg *Vj = (VReg *)vj; \
+ Vd->E(imm & MASK) = Vj->E(0); \
+}
+
+XVINSVE0(xvinsve0_w, W, 0x7)
+XVINSVE0(xvinsve0_d, D, 0x3)
+
+#define XVPICKVE(NAME, E, BIT, MASK) \
+void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
+{ \
+ int i; \
+ VReg *Vd = (VReg *)vd; \
+ VReg *Vj = (VReg *)vj; \
+ int oprsz = simd_oprsz(desc); \
+ \
+ Vd->E(0) = Vj->E(imm & MASK); \
+ for (i = 1; i < oprsz / (BIT / 8); i++) { \
+ Vd->E(i) = 0; \
+ } \
+}
+
+XVPICKVE(xvpickve_w, W, 32, 0x7)
+XVPICKVE(xvpickve_d, D, 64, 0x3)
+
#define VPACKEV(NAME, BIT, E) \
void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
{ \
