target/loongarch: Implement vssrlrn vssrarn
This patch includes:
- VSSRLRN.{B.H/H.W/W.D};
- VSSRARN.{B.H/H.W/W.D};
- VSSRLRN.{BU.H/HU.W/WU.D};
- VSSRARN.{BU.H/HU.W/WU.D};
- VSSRLRNI.{B.H/H.W/W.D/D.Q};
- VSSRARNI.{B.H/H.W/W.D/D.Q};
- VSSRLRNI.{BU.H/HU.W/WU.D/DU.Q};
- VSSRARNI.{BU.H/HU.W/WU.D/DU.Q}.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Song Gao <gaosong@loongson.cn>
Message-Id: <20230504122810.4094787-29-gaosong@loongson.cn>
diff --git a/target/loongarch/lsx_helper.c b/target/loongarch/lsx_helper.c
index b9110dc..5fc22ea 100644
--- a/target/loongarch/lsx_helper.c
+++ b/target/loongarch/lsx_helper.c
@@ -1557,3 +1557,361 @@
VSSRANUI(vssrani_bu_h, 16, B, H)
VSSRANUI(vssrani_hu_w, 32, H, W)
VSSRANUI(vssrani_wu_d, 64, W, D)
+
+#define SSRLRNS(E1, E2, T1, T2, T3) \
+static T1 do_ssrlrns_ ## E1(T2 e2, int sa, int sh) \
+{ \
+ T1 shft_res; \
+ \
+ shft_res = do_vsrlr_ ## E2(e2, sa); \
+ T1 mask; \
+ mask = (1ull << sh) -1; \
+ if (shft_res > mask) { \
+ return mask; \
+ } else { \
+ return shft_res; \
+ } \
+}
+
+SSRLRNS(B, H, uint16_t, int16_t, uint8_t)
+SSRLRNS(H, W, uint32_t, int32_t, uint16_t)
+SSRLRNS(W, D, uint64_t, int64_t, uint32_t)
+
+#define VSSRLRN(NAME, BIT, T, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t vk) \
+{ \
+ int i; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ VReg *Vk = &(env->fpr[vk].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ Vd->E1(i) = do_ssrlrns_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2 -1); \
+ } \
+ Vd->D(1) = 0; \
+}
+
+VSSRLRN(vssrlrn_b_h, 16, uint16_t, B, H)
+VSSRLRN(vssrlrn_h_w, 32, uint32_t, H, W)
+VSSRLRN(vssrlrn_w_d, 64, uint64_t, W, D)
+
+#define SSRARNS(E1, E2, T1, T2) \
+static T1 do_ssrarns_ ## E1(T1 e2, int sa, int sh) \
+{ \
+ T1 shft_res; \
+ \
+ shft_res = do_vsrar_ ## E2(e2, sa); \
+ T2 mask; \
+ mask = (1ll << sh) -1; \
+ if (shft_res > mask) { \
+ return mask; \
+ } else if (shft_res < -(mask +1)) { \
+ return ~mask; \
+ } else { \
+ return shft_res; \
+ } \
+}
+
+SSRARNS(B, H, int16_t, int8_t)
+SSRARNS(H, W, int32_t, int16_t)
+SSRARNS(W, D, int64_t, int32_t)
+
+#define VSSRARN(NAME, BIT, T, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t vk) \
+{ \
+ int i; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ VReg *Vk = &(env->fpr[vk].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ Vd->E1(i) = do_ssrarns_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2 -1); \
+ } \
+ Vd->D(1) = 0; \
+}
+
+VSSRARN(vssrarn_b_h, 16, uint16_t, B, H)
+VSSRARN(vssrarn_h_w, 32, uint32_t, H, W)
+VSSRARN(vssrarn_w_d, 64, uint64_t, W, D)
+
+#define SSRLRNU(E1, E2, T1, T2, T3) \
+static T1 do_ssrlrnu_ ## E1(T3 e2, int sa, int sh) \
+{ \
+ T1 shft_res; \
+ \
+ shft_res = do_vsrlr_ ## E2(e2, sa); \
+ \
+ T2 mask; \
+ mask = (1ull << sh) -1; \
+ if (shft_res > mask) { \
+ return mask; \
+ } else { \
+ return shft_res; \
+ } \
+}
+
+SSRLRNU(B, H, uint16_t, uint8_t, int16_t)
+SSRLRNU(H, W, uint32_t, uint16_t, int32_t)
+SSRLRNU(W, D, uint64_t, uint32_t, int64_t)
+
+#define VSSRLRNU(NAME, BIT, T, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t vk) \
+{ \
+ int i; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ VReg *Vk = &(env->fpr[vk].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ Vd->E1(i) = do_ssrlrnu_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2); \
+ } \
+ Vd->D(1) = 0; \
+}
+
+VSSRLRNU(vssrlrn_bu_h, 16, uint16_t, B, H)
+VSSRLRNU(vssrlrn_hu_w, 32, uint32_t, H, W)
+VSSRLRNU(vssrlrn_wu_d, 64, uint64_t, W, D)
+
+#define SSRARNU(E1, E2, T1, T2, T3) \
+static T1 do_ssrarnu_ ## E1(T3 e2, int sa, int sh) \
+{ \
+ T1 shft_res; \
+ \
+ if (e2 < 0) { \
+ shft_res = 0; \
+ } else { \
+ shft_res = do_vsrar_ ## E2(e2, sa); \
+ } \
+ T2 mask; \
+ mask = (1ull << sh) -1; \
+ if (shft_res > mask) { \
+ return mask; \
+ } else { \
+ return shft_res; \
+ } \
+}
+
+SSRARNU(B, H, uint16_t, uint8_t, int16_t)
+SSRARNU(H, W, uint32_t, uint16_t, int32_t)
+SSRARNU(W, D, uint64_t, uint32_t, int64_t)
+
+#define VSSRARNU(NAME, BIT, T, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t vk) \
+{ \
+ int i; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ VReg *Vk = &(env->fpr[vk].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ Vd->E1(i) = do_ssrarnu_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2); \
+ } \
+ Vd->D(1) = 0; \
+}
+
+VSSRARNU(vssrarn_bu_h, 16, uint16_t, B, H)
+VSSRARNU(vssrarn_hu_w, 32, uint32_t, H, W)
+VSSRARNU(vssrarn_wu_d, 64, uint64_t, W, D)
+
+#define VSSRLRNI(NAME, BIT, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t imm) \
+{ \
+ int i; \
+ VReg temp; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ temp.E1(i) = do_ssrlrns_ ## E1(Vj->E2(i), imm, BIT/2 -1); \
+ temp.E1(i + LSX_LEN/BIT) = do_ssrlrns_ ## E1(Vd->E2(i), imm, BIT/2 -1);\
+ } \
+ *Vd = temp; \
+}
+
+#define VSSRLRNI_Q(NAME, sh) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t imm) \
+{ \
+ Int128 shft_res1, shft_res2, mask, r1, r2; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ \
+ if (imm == 0) { \
+ shft_res1 = Vj->Q(0); \
+ shft_res2 = Vd->Q(0); \
+ } else { \
+ r1 = int128_and(int128_urshift(Vj->Q(0), (imm -1)), int128_one()); \
+ r2 = int128_and(int128_urshift(Vd->Q(0), (imm -1)), int128_one()); \
+ \
+ shft_res1 = (int128_add(int128_urshift(Vj->Q(0), imm), r1)); \
+ shft_res2 = (int128_add(int128_urshift(Vd->Q(0), imm), r2)); \
+ } \
+ \
+ mask = int128_sub(int128_lshift(int128_one(), sh), int128_one()); \
+ \
+ if (int128_ult(mask, shft_res1)) { \
+ Vd->D(0) = int128_getlo(mask); \
+ }else { \
+ Vd->D(0) = int128_getlo(shft_res1); \
+ } \
+ \
+ if (int128_ult(mask, shft_res2)) { \
+ Vd->D(1) = int128_getlo(mask); \
+ }else { \
+ Vd->D(1) = int128_getlo(shft_res2); \
+ } \
+}
+
+VSSRLRNI(vssrlrni_b_h, 16, B, H)
+VSSRLRNI(vssrlrni_h_w, 32, H, W)
+VSSRLRNI(vssrlrni_w_d, 64, W, D)
+VSSRLRNI_Q(vssrlrni_d_q, 63)
+
+#define VSSRARNI(NAME, BIT, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t imm) \
+{ \
+ int i; \
+ VReg temp; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ temp.E1(i) = do_ssrarns_ ## E1(Vj->E2(i), imm, BIT/2 -1); \
+ temp.E1(i + LSX_LEN/BIT) = do_ssrarns_ ## E1(Vd->E2(i), imm, BIT/2 -1); \
+ } \
+ *Vd = temp; \
+}
+
+void HELPER(vssrarni_d_q)(CPULoongArchState *env,
+ uint32_t vd, uint32_t vj, uint32_t imm)
+{
+ Int128 shft_res1, shft_res2, mask1, mask2, r1, r2;
+ VReg *Vd = &(env->fpr[vd].vreg);
+ VReg *Vj = &(env->fpr[vj].vreg);
+
+ if (imm == 0) {
+ shft_res1 = Vj->Q(0);
+ shft_res2 = Vd->Q(0);
+ } else {
+ r1 = int128_and(int128_rshift(Vj->Q(0), (imm -1)), int128_one());
+ r2 = int128_and(int128_rshift(Vd->Q(0), (imm -1)), int128_one());
+
+ shft_res1 = int128_add(int128_rshift(Vj->Q(0), imm), r1);
+ shft_res2 = int128_add(int128_rshift(Vd->Q(0), imm), r2);
+ }
+
+ mask1 = int128_sub(int128_lshift(int128_one(), 63), int128_one());
+ mask2 = int128_lshift(int128_one(), 63);
+
+ if (int128_gt(shft_res1, mask1)) {
+ Vd->D(0) = int128_getlo(mask1);
+ } else if (int128_lt(shft_res1, int128_neg(mask2))) {
+ Vd->D(0) = int128_getlo(mask2);
+ } else {
+ Vd->D(0) = int128_getlo(shft_res1);
+ }
+
+ if (int128_gt(shft_res2, mask1)) {
+ Vd->D(1) = int128_getlo(mask1);
+ } else if (int128_lt(shft_res2, int128_neg(mask2))) {
+ Vd->D(1) = int128_getlo(mask2);
+ } else {
+ Vd->D(1) = int128_getlo(shft_res2);
+ }
+}
+
+VSSRARNI(vssrarni_b_h, 16, B, H)
+VSSRARNI(vssrarni_h_w, 32, H, W)
+VSSRARNI(vssrarni_w_d, 64, W, D)
+
+#define VSSRLRNUI(NAME, BIT, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t imm) \
+{ \
+ int i; \
+ VReg temp; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ temp.E1(i) = do_ssrlrnu_ ## E1(Vj->E2(i), imm, BIT/2); \
+ temp.E1(i + LSX_LEN/BIT) = do_ssrlrnu_ ## E1(Vd->E2(i), imm, BIT/2); \
+ } \
+ *Vd = temp; \
+}
+
+VSSRLRNUI(vssrlrni_bu_h, 16, B, H)
+VSSRLRNUI(vssrlrni_hu_w, 32, H, W)
+VSSRLRNUI(vssrlrni_wu_d, 64, W, D)
+VSSRLRNI_Q(vssrlrni_du_q, 64)
+
+#define VSSRARNUI(NAME, BIT, E1, E2) \
+void HELPER(NAME)(CPULoongArchState *env, \
+ uint32_t vd, uint32_t vj, uint32_t imm) \
+{ \
+ int i; \
+ VReg temp; \
+ VReg *Vd = &(env->fpr[vd].vreg); \
+ VReg *Vj = &(env->fpr[vj].vreg); \
+ \
+ for (i = 0; i < LSX_LEN/BIT; i++) { \
+ temp.E1(i) = do_ssrarnu_ ## E1(Vj->E2(i), imm, BIT/2); \
+ temp.E1(i + LSX_LEN/BIT) = do_ssrarnu_ ## E1(Vd->E2(i), imm, BIT/2); \
+ } \
+ *Vd = temp; \
+}
+
+void HELPER(vssrarni_du_q)(CPULoongArchState *env,
+ uint32_t vd, uint32_t vj, uint32_t imm)
+{
+ Int128 shft_res1, shft_res2, mask1, mask2, r1, r2;
+ VReg *Vd = &(env->fpr[vd].vreg);
+ VReg *Vj = &(env->fpr[vj].vreg);
+
+ if (imm == 0) {
+ shft_res1 = Vj->Q(0);
+ shft_res2 = Vd->Q(0);
+ } else {
+ r1 = int128_and(int128_rshift(Vj->Q(0), (imm -1)), int128_one());
+ r2 = int128_and(int128_rshift(Vd->Q(0), (imm -1)), int128_one());
+
+ shft_res1 = int128_add(int128_rshift(Vj->Q(0), imm), r1);
+ shft_res2 = int128_add(int128_rshift(Vd->Q(0), imm), r2);
+ }
+
+ if (int128_lt(Vj->Q(0), int128_zero())) {
+ shft_res1 = int128_zero();
+ }
+ if (int128_lt(Vd->Q(0), int128_zero())) {
+ shft_res2 = int128_zero();
+ }
+
+ mask1 = int128_sub(int128_lshift(int128_one(), 64), int128_one());
+ mask2 = int128_lshift(int128_one(), 64);
+
+ if (int128_gt(shft_res1, mask1)) {
+ Vd->D(0) = int128_getlo(mask1);
+ } else if (int128_lt(shft_res1, int128_neg(mask2))) {
+ Vd->D(0) = int128_getlo(mask2);
+ } else {
+ Vd->D(0) = int128_getlo(shft_res1);
+ }
+
+ if (int128_gt(shft_res2, mask1)) {
+ Vd->D(1) = int128_getlo(mask1);
+ } else if (int128_lt(shft_res2, int128_neg(mask2))) {
+ Vd->D(1) = int128_getlo(mask2);
+ } else {
+ Vd->D(1) = int128_getlo(shft_res2);
+ }
+}
+
+VSSRARNUI(vssrarni_bu_h, 16, B, H)
+VSSRARNUI(vssrarni_hu_w, 32, H, W)
+VSSRARNUI(vssrarni_wu_d, 64, W, D)
