diff --git a/target/arm/helper.h b/target/arm/helper.h index f830531dd3..de2c5c9aef 100644 --- a/target/arm/helper.h +++ b/target/arm/helper.h @@ -836,6 +836,22 @@ DEF_HELPER_FLAGS_5(gvec_sqsub_s, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(gvec_sqsub_d, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_usqadd_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_usqadd_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_usqadd_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_usqadd_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_suqadd_b, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_suqadd_h, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_suqadd_s, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) +DEF_HELPER_FLAGS_5(gvec_suqadd_d, TCG_CALL_NO_RWG, + void, ptr, ptr, ptr, ptr, i32) DEF_HELPER_FLAGS_5(gvec_fmlal_a32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32) diff --git a/target/arm/tcg/gengvec64.c b/target/arm/tcg/gengvec64.c index 093b498b13..b3afabd38b 100644 --- a/target/arm/tcg/gengvec64.c +++ b/target/arm/tcg/gengvec64.c @@ -188,3 +188,113 @@ void gen_gvec_bcax(unsigned vece, uint32_t d, uint32_t n, uint32_t m, tcg_gen_gvec_4(d, n, m, a, oprsz, maxsz, &op); } +static void gen_suqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec qc, + TCGv_vec a, TCGv_vec b) +{ + TCGv_vec max = + tcg_constant_vec_matching(t, vece, (1ull << ((8 << vece) - 1)) - 1); + TCGv_vec u = tcg_temp_new_vec_matching(t); + + /* Maximum value that can be added to @a without overflow. */ + tcg_gen_sub_vec(vece, u, max, a); + + /* Constrain addend so that the next addition never overflows. */ + tcg_gen_umin_vec(vece, u, u, b); + tcg_gen_add_vec(vece, t, u, a); + + /* Compute QC by comparing the adjusted @b. */ + tcg_gen_xor_vec(vece, u, u, b); + tcg_gen_or_vec(vece, qc, qc, u); +} + +void gen_gvec_suqadd_qc(unsigned vece, uint32_t rd_ofs, + uint32_t rn_ofs, uint32_t rm_ofs, + uint32_t opr_sz, uint32_t max_sz) +{ + static const TCGOpcode vecop_list[] = { + INDEX_op_add_vec, INDEX_op_sub_vec, INDEX_op_umin_vec, 0 + }; + static const GVecGen4 ops[4] = { + { .fniv = gen_suqadd_vec, + .fno = gen_helper_gvec_suqadd_b, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_8 }, + { .fniv = gen_suqadd_vec, + .fno = gen_helper_gvec_suqadd_h, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_16 }, + { .fniv = gen_suqadd_vec, + .fno = gen_helper_gvec_suqadd_s, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_32 }, + { .fniv = gen_suqadd_vec, + .fno = gen_helper_gvec_suqadd_d, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_64 }, + }; + + tcg_debug_assert(opr_sz <= sizeof_field(CPUARMState, vfp.qc)); + tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc), + rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]); +} + +static void gen_usqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec qc, + TCGv_vec a, TCGv_vec b) +{ + TCGv_vec u = tcg_temp_new_vec_matching(t); + TCGv_vec z = tcg_constant_vec_matching(t, vece, 0); + + /* Compute unsigned saturation of add for +b and sub for -b. */ + tcg_gen_neg_vec(vece, t, b); + tcg_gen_usadd_vec(vece, u, a, b); + tcg_gen_ussub_vec(vece, t, a, t); + + /* Select the correct result depending on the sign of b. */ + tcg_gen_cmpsel_vec(TCG_COND_LT, vece, t, b, z, t, u); + + /* Compute QC by comparing against the non-saturated result. */ + tcg_gen_add_vec(vece, u, a, b); + tcg_gen_xor_vec(vece, u, u, t); + tcg_gen_or_vec(vece, qc, qc, u); +} + +void gen_gvec_usqadd_qc(unsigned vece, uint32_t rd_ofs, + uint32_t rn_ofs, uint32_t rm_ofs, + uint32_t opr_sz, uint32_t max_sz) +{ + static const TCGOpcode vecop_list[] = { + INDEX_op_neg_vec, INDEX_op_add_vec, + INDEX_op_usadd_vec, INDEX_op_ussub_vec, + INDEX_op_cmpsel_vec, 0 + }; + static const GVecGen4 ops[4] = { + { .fniv = gen_usqadd_vec, + .fno = gen_helper_gvec_usqadd_b, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_8 }, + { .fniv = gen_usqadd_vec, + .fno = gen_helper_gvec_usqadd_h, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_16 }, + { .fniv = gen_usqadd_vec, + .fno = gen_helper_gvec_usqadd_s, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_32 }, + { .fniv = gen_usqadd_vec, + .fno = gen_helper_gvec_usqadd_d, + .opt_opc = vecop_list, + .write_aofs = true, + .vece = MO_64 }, + }; + + tcg_debug_assert(opr_sz <= sizeof_field(CPUARMState, vfp.qc)); + tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc), + rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]); +} diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c index 9167e4d0bd..9f948e033e 100644 --- a/target/arm/tcg/translate-a64.c +++ b/target/arm/tcg/translate-a64.c @@ -9983,83 +9983,68 @@ static void handle_2misc_narrow(DisasContext *s, bool scalar, /* Remaining saturating accumulating ops */ static void handle_2misc_satacc(DisasContext *s, bool is_scalar, bool is_u, - bool is_q, int size, int rn, int rd) + bool is_q, unsigned size, int rn, int rd) { - bool is_double = (size == 3); + if (!is_scalar) { + gen_gvec_fn3(s, is_q, rd, rd, rn, + is_u ? gen_gvec_usqadd_qc : gen_gvec_suqadd_qc, size); + return; + } - if (is_double) { + if (size == 3) { TCGv_i64 tcg_rn = tcg_temp_new_i64(); TCGv_i64 tcg_rd = tcg_temp_new_i64(); - int pass; - for (pass = 0; pass < (is_scalar ? 1 : 2); pass++) { - read_vec_element(s, tcg_rn, rn, pass, MO_64); - read_vec_element(s, tcg_rd, rd, pass, MO_64); + read_vec_element(s, tcg_rn, rn, 0, MO_64); + read_vec_element(s, tcg_rd, rd, 0, MO_64); - if (is_u) { /* USQADD */ - gen_helper_neon_uqadd_s64(tcg_rd, tcg_env, tcg_rn, tcg_rd); - } else { /* SUQADD */ - gen_helper_neon_sqadd_u64(tcg_rd, tcg_env, tcg_rn, tcg_rd); - } - write_vec_element(s, tcg_rd, rd, pass, MO_64); + if (is_u) { /* USQADD */ + gen_helper_neon_uqadd_s64(tcg_rd, tcg_env, tcg_rn, tcg_rd); + } else { /* SUQADD */ + gen_helper_neon_sqadd_u64(tcg_rd, tcg_env, tcg_rn, tcg_rd); } - clear_vec_high(s, !is_scalar, rd); + write_vec_element(s, tcg_rd, rd, 0, MO_64); + clear_vec_high(s, false, rd); } else { TCGv_i32 tcg_rn = tcg_temp_new_i32(); TCGv_i32 tcg_rd = tcg_temp_new_i32(); - int pass, maxpasses; - if (is_scalar) { - maxpasses = 1; - } else { - maxpasses = is_q ? 4 : 2; + read_vec_element_i32(s, tcg_rn, rn, 0, size); + read_vec_element_i32(s, tcg_rd, rd, 0, size); + + if (is_u) { /* USQADD */ + switch (size) { + case 0: + gen_helper_neon_uqadd_s8(tcg_rd, tcg_env, tcg_rn, tcg_rd); + break; + case 1: + gen_helper_neon_uqadd_s16(tcg_rd, tcg_env, tcg_rn, tcg_rd); + break; + case 2: + gen_helper_neon_uqadd_s32(tcg_rd, tcg_env, tcg_rn, tcg_rd); + break; + default: + g_assert_not_reached(); + } + } else { /* SUQADD */ + switch (size) { + case 0: + gen_helper_neon_sqadd_u8(tcg_rd, tcg_env, tcg_rn, tcg_rd); + break; + case 1: + gen_helper_neon_sqadd_u16(tcg_rd, tcg_env, tcg_rn, tcg_rd); + break; + case 2: + gen_helper_neon_sqadd_u32(tcg_rd, tcg_env, tcg_rn, tcg_rd); + break; + default: + g_assert_not_reached(); + } } - for (pass = 0; pass < maxpasses; pass++) { - if (is_scalar) { - read_vec_element_i32(s, tcg_rn, rn, pass, size); - read_vec_element_i32(s, tcg_rd, rd, pass, size); - } else { - read_vec_element_i32(s, tcg_rn, rn, pass, MO_32); - read_vec_element_i32(s, tcg_rd, rd, pass, MO_32); - } - - if (is_u) { /* USQADD */ - switch (size) { - case 0: - gen_helper_neon_uqadd_s8(tcg_rd, tcg_env, tcg_rn, tcg_rd); - break; - case 1: - gen_helper_neon_uqadd_s16(tcg_rd, tcg_env, tcg_rn, tcg_rd); - break; - case 2: - gen_helper_neon_uqadd_s32(tcg_rd, tcg_env, tcg_rn, tcg_rd); - break; - default: - g_assert_not_reached(); - } - } else { /* SUQADD */ - switch (size) { - case 0: - gen_helper_neon_sqadd_u8(tcg_rd, tcg_env, tcg_rn, tcg_rd); - break; - case 1: - gen_helper_neon_sqadd_u16(tcg_rd, tcg_env, tcg_rn, tcg_rd); - break; - case 2: - gen_helper_neon_sqadd_u32(tcg_rd, tcg_env, tcg_rn, tcg_rd); - break; - default: - g_assert_not_reached(); - } - } - - if (is_scalar) { - write_vec_element(s, tcg_constant_i64(0), rd, 0, MO_64); - } - write_vec_element_i32(s, tcg_rd, rd, pass, MO_32); - } - clear_vec_high(s, is_q, rd); + write_vec_element(s, tcg_constant_i64(0), rd, 0, MO_64); + write_vec_element_i32(s, tcg_rd, rd, 0, MO_32); + clear_vec_high(s, false, rd); } } diff --git a/target/arm/tcg/translate-a64.h b/target/arm/tcg/translate-a64.h index 91750f0ca9..b5cb26f8a2 100644 --- a/target/arm/tcg/translate-a64.h +++ b/target/arm/tcg/translate-a64.h @@ -197,6 +197,12 @@ void gen_gvec_eor3(unsigned vece, uint32_t d, uint32_t n, uint32_t m, uint32_t a, uint32_t oprsz, uint32_t maxsz); void gen_gvec_bcax(unsigned vece, uint32_t d, uint32_t n, uint32_t m, uint32_t a, uint32_t oprsz, uint32_t maxsz); +void gen_gvec_suqadd_qc(unsigned vece, uint32_t rd_ofs, + uint32_t rn_ofs, uint32_t rm_ofs, + uint32_t opr_sz, uint32_t max_sz); +void gen_gvec_usqadd_qc(unsigned vece, uint32_t rd_ofs, + uint32_t rn_ofs, uint32_t rm_ofs, + uint32_t opr_sz, uint32_t max_sz); void gen_sve_ldr(DisasContext *s, TCGv_ptr, int vofs, int len, int rn, int imm); void gen_sve_str(DisasContext *s, TCGv_ptr, int vofs, int len, int rn, int imm); diff --git a/target/arm/tcg/vec_helper.c b/target/arm/tcg/vec_helper.c index 56fea14edb..d8e96386be 100644 --- a/target/arm/tcg/vec_helper.c +++ b/target/arm/tcg/vec_helper.c @@ -1555,6 +1555,14 @@ DO_SAT(gvec_sqsub_b, int, int8_t, int8_t, -, INT8_MIN, INT8_MAX) DO_SAT(gvec_sqsub_h, int, int16_t, int16_t, -, INT16_MIN, INT16_MAX) DO_SAT(gvec_sqsub_s, int64_t, int32_t, int32_t, -, INT32_MIN, INT32_MAX) +DO_SAT(gvec_usqadd_b, int, uint8_t, int8_t, +, 0, UINT8_MAX) +DO_SAT(gvec_usqadd_h, int, uint16_t, int16_t, +, 0, UINT16_MAX) +DO_SAT(gvec_usqadd_s, int64_t, uint32_t, int32_t, +, 0, UINT32_MAX) + +DO_SAT(gvec_suqadd_b, int, int8_t, uint8_t, +, INT8_MIN, INT8_MAX) +DO_SAT(gvec_suqadd_h, int, int16_t, uint16_t, +, INT16_MIN, INT16_MAX) +DO_SAT(gvec_suqadd_s, int64_t, int32_t, uint32_t, +, INT32_MIN, INT32_MAX) + #undef DO_SAT void HELPER(gvec_uqadd_d)(void *vd, void *vq, void *vn, @@ -1645,6 +1653,62 @@ void HELPER(gvec_sqsub_d)(void *vd, void *vq, void *vn, clear_tail(d, oprsz, simd_maxsz(desc)); } +void HELPER(gvec_usqadd_d)(void *vd, void *vq, void *vn, + void *vm, uint32_t desc) +{ + intptr_t i, oprsz = simd_oprsz(desc); + uint64_t *d = vd, *n = vn, *m = vm; + bool q = false; + + for (i = 0; i < oprsz / 8; i++) { + uint64_t nn = n[i]; + int64_t mm = m[i]; + uint64_t dd = nn + mm; + + if (mm < 0) { + if (nn < (uint64_t)-mm) { + dd = 0; + q = true; + } + } else { + if (dd < nn) { + dd = UINT64_MAX; + q = true; + } + } + d[i] = dd; + } + if (q) { + uint32_t *qc = vq; + qc[0] = 1; + } + clear_tail(d, oprsz, simd_maxsz(desc)); +} + +void HELPER(gvec_suqadd_d)(void *vd, void *vq, void *vn, + void *vm, uint32_t desc) +{ + intptr_t i, oprsz = simd_oprsz(desc); + uint64_t *d = vd, *n = vn, *m = vm; + bool q = false; + + for (i = 0; i < oprsz / 8; i++) { + int64_t nn = n[i]; + uint64_t mm = m[i]; + int64_t dd = nn + mm; + + if (mm > (uint64_t)(INT64_MAX - nn)) { + dd = INT64_MAX; + q = true; + } + d[i] = dd; + } + if (q) { + uint32_t *qc = vq; + qc[0] = 1; + } + clear_tail(d, oprsz, simd_maxsz(desc)); +} #define DO_SRA(NAME, TYPE) \ void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \