target/arm: Implement ARMv8.5-FRINT

Tested-by: NLaurent Desnogues <laurent.desnogues@gmail.com>
Signed-off-by: NRichard Henderson <richard.henderson@linaro.org>
Message-id: 20190301200501.16533-11-richard.henderson@linaro.org
Reviewed-by: NPeter Maydell <peter.maydell@linaro.org>
Signed-off-by: NPeter Maydell <peter.maydell@linaro.org>

target/arm/cpu.h

@@ -3476,6 +3476,11 @@ static inline bool isar_feature_aa64_predinv(const ARMISARegisters *id)
     return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, SPECRES) != 0;
 }
 
+static inline bool isar_feature_aa64_frint(const ARMISARegisters *id)
+{
+    return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, FRINTTS) != 0;
+}
+
 static inline bool isar_feature_aa64_fp16(const ARMISARegisters *id)
 {
     /* We always set the AdvSIMD and FP fields identically wrt FP16.  */

target/arm/cpu64.c

@@ -321,6 +321,7 @@ static void aarch64_max_initfn(Object *obj)
         t = FIELD_DP64(t, ID_AA64ISAR1, GPI, 0);
         t = FIELD_DP64(t, ID_AA64ISAR1, SB, 1);
         t = FIELD_DP64(t, ID_AA64ISAR1, SPECRES, 1);
+        t = FIELD_DP64(t, ID_AA64ISAR1, FRINTTS, 1);
         cpu->isar.id_aa64isar1 = t;
 
         t = cpu->isar.id_aa64pfr0;

target/arm/helper.h

@@ -683,6 +683,11 @@ DEF_HELPER_FLAGS_5(gvec_fmlal_idx_a32, TCG_CALL_NO_RWG,
 DEF_HELPER_FLAGS_5(gvec_fmlal_idx_a64, TCG_CALL_NO_RWG,
                    void, ptr, ptr, ptr, ptr, i32)
 
+DEF_HELPER_FLAGS_2(frint32_s, TCG_CALL_NO_RWG, f32, f32, ptr)
+DEF_HELPER_FLAGS_2(frint64_s, TCG_CALL_NO_RWG, f32, f32, ptr)
+DEF_HELPER_FLAGS_2(frint32_d, TCG_CALL_NO_RWG, f64, f64, ptr)
+DEF_HELPER_FLAGS_2(frint64_d, TCG_CALL_NO_RWG, f64, f64, ptr)
+
 #ifdef TARGET_AARCH64
 #include "helper-a64.h"
 #include "helper-sve.h"

target/arm/translate-a64.c

@@ -5723,6 +5723,20 @@ static void handle_fp_1src_single(DisasContext *s, int opcode, int rd, int rn)
     case 0xf: /* FRINTI */
         gen_fpst = gen_helper_rints;
         break;
+    case 0x10: /* FRINT32Z */
+        rmode = float_round_to_zero;
+        gen_fpst = gen_helper_frint32_s;
+        break;
+    case 0x11: /* FRINT32X */
+        gen_fpst = gen_helper_frint32_s;
+        break;
+    case 0x12: /* FRINT64Z */
+        rmode = float_round_to_zero;
+        gen_fpst = gen_helper_frint64_s;
+        break;
+    case 0x13: /* FRINT64X */
+        gen_fpst = gen_helper_frint64_s;
+        break;
     default:
         g_assert_not_reached();
     }
@@ -5786,6 +5800,20 @@ static void handle_fp_1src_double(DisasContext *s, int opcode, int rd, int rn)
     case 0xf: /* FRINTI */
         gen_fpst = gen_helper_rintd;
         break;
+    case 0x10: /* FRINT32Z */
+        rmode = float_round_to_zero;
+        gen_fpst = gen_helper_frint32_d;
+        break;
+    case 0x11: /* FRINT32X */
+        gen_fpst = gen_helper_frint32_d;
+        break;
+    case 0x12: /* FRINT64Z */
+        rmode = float_round_to_zero;
+        gen_fpst = gen_helper_frint64_d;
+        break;
+    case 0x13: /* FRINT64X */
+        gen_fpst = gen_helper_frint64_d;
+        break;
     default:
         g_assert_not_reached();
     }
@@ -5922,6 +5950,13 @@ static void disas_fp_1src(DisasContext *s, uint32_t insn)
         handle_fp_fcvt(s, opcode, rd, rn, dtype, type);
         break;
     }
+
+    case 0x10 ... 0x13: /* FRINT{32,64}{X,Z} */
+        if (type > 1 || !dc_isar_feature(aa64_frint, s)) {
+            unallocated_encoding(s);
+            return;
+        }
+        /* fall through */
     case 0x0 ... 0x3:
     case 0x8 ... 0xc:
     case 0xe ... 0xf:
@@ -5931,14 +5966,12 @@ static void disas_fp_1src(DisasContext *s, uint32_t insn)
             if (!fp_access_check(s)) {
                 return;
             }
-
             handle_fp_1src_single(s, opcode, rd, rn);
             break;
         case 1:
             if (!fp_access_check(s)) {
                 return;
             }
-
             handle_fp_1src_double(s, opcode, rd, rn);
             break;
         case 3:
@@ -5950,13 +5983,13 @@ static void disas_fp_1src(DisasContext *s, uint32_t insn)
             if (!fp_access_check(s)) {
                 return;
             }
-
             handle_fp_1src_half(s, opcode, rd, rn);
             break;
         default:
             unallocated_encoding(s);
         }
         break;
+
     default:
         unallocated_encoding(s);
         break;
@@ -9484,6 +9517,14 @@ static void handle_2misc_64(DisasContext *s, int opcode, bool u,
     case 0x59: /* FRINTX */
         gen_helper_rintd_exact(tcg_rd, tcg_rn, tcg_fpstatus);
         break;
+    case 0x1e: /* FRINT32Z */
+    case 0x5e: /* FRINT32X */
+        gen_helper_frint32_d(tcg_rd, tcg_rn, tcg_fpstatus);
+        break;
+    case 0x1f: /* FRINT64Z */
+    case 0x5f: /* FRINT64X */
+        gen_helper_frint64_d(tcg_rd, tcg_rn, tcg_fpstatus);
+        break;
     default:
         g_assert_not_reached();
     }
@@ -12134,8 +12175,7 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
         }
         break;
     case 0xc ... 0xf:
-    case 0x16 ... 0x1d:
-    case 0x1f:
+    case 0x16 ... 0x1f:
     {
         /* Floating point: U, size[1] and opcode indicate operation;
          * size[0] indicates single or double precision.
@@ -12278,6 +12318,19 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
             }
             need_fpstatus = true;
             break;
+        case 0x1e: /* FRINT32Z */
+        case 0x1f: /* FRINT64Z */
+            need_rmode = true;
+            rmode = FPROUNDING_ZERO;
+            /* fall through */
+        case 0x5e: /* FRINT32X */
+        case 0x5f: /* FRINT64X */
+            need_fpstatus = true;
+            if ((size == 3 && !is_q) || !dc_isar_feature(aa64_frint, s)) {
+                unallocated_encoding(s);
+                return;
+            }
+            break;
         default:
             unallocated_encoding(s);
             return;
@@ -12443,6 +12496,14 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
                 case 0x7c: /* URSQRTE */
                     gen_helper_rsqrte_u32(tcg_res, tcg_op, tcg_fpstatus);
                     break;
+                case 0x1e: /* FRINT32Z */
+                case 0x5e: /* FRINT32X */
+                    gen_helper_frint32_s(tcg_res, tcg_op, tcg_fpstatus);
+                    break;
+                case 0x1f: /* FRINT64Z */
+                case 0x5f: /* FRINT64X */
+                    gen_helper_frint64_s(tcg_res, tcg_op, tcg_fpstatus);
+                    break;
                 default:
                     g_assert_not_reached();
                 }

target/arm/vfp_helper.c

@@ -1174,3 +1174,99 @@ uint32_t HELPER(vjcvt)(float64 value, CPUARMState *env)
 
     return result;
 }
+
+/* Round a float32 to an integer that fits in int32_t or int64_t.  */
+static float32 frint_s(float32 f, float_status *fpst, int intsize)
+{
+    int old_flags = get_float_exception_flags(fpst);
+    uint32_t exp = extract32(f, 23, 8);
+
+    if (unlikely(exp == 0xff)) {
+        /* NaN or Inf.  */
+        goto overflow;
+    }
+
+    /* Round and re-extract the exponent.  */
+    f = float32_round_to_int(f, fpst);
+    exp = extract32(f, 23, 8);
+
+    /* Validate the range of the result.  */
+    if (exp < 126 + intsize) {
+        /* abs(F) <= INT{N}_MAX */
+        return f;
+    }
+    if (exp == 126 + intsize) {
+        uint32_t sign = extract32(f, 31, 1);
+        uint32_t frac = extract32(f, 0, 23);
+        if (sign && frac == 0) {
+            /* F == INT{N}_MIN */
+            return f;
+        }
+    }
+
+ overflow:
+    /*
+     * Raise Invalid and return INT{N}_MIN as a float.  Revert any
+     * inexact exception float32_round_to_int may have raised.
+     */
+    set_float_exception_flags(old_flags | float_flag_invalid, fpst);
+    return (0x100u + 126u + intsize) << 23;
+}
+
+float32 HELPER(frint32_s)(float32 f, void *fpst)
+{
+    return frint_s(f, fpst, 32);
+}
+
+float32 HELPER(frint64_s)(float32 f, void *fpst)
+{
+    return frint_s(f, fpst, 64);
+}
+
+/* Round a float64 to an integer that fits in int32_t or int64_t.  */
+static float64 frint_d(float64 f, float_status *fpst, int intsize)
+{
+    int old_flags = get_float_exception_flags(fpst);
+    uint32_t exp = extract64(f, 52, 11);
+
+    if (unlikely(exp == 0x7ff)) {
+        /* NaN or Inf.  */
+        goto overflow;
+    }
+
+    /* Round and re-extract the exponent.  */
+    f = float64_round_to_int(f, fpst);
+    exp = extract64(f, 52, 11);
+
+    /* Validate the range of the result.  */
+    if (exp < 1022 + intsize) {
+        /* abs(F) <= INT{N}_MAX */
+        return f;
+    }
+    if (exp == 1022 + intsize) {
+        uint64_t sign = extract64(f, 63, 1);
+        uint64_t frac = extract64(f, 0, 52);
+        if (sign && frac == 0) {
+            /* F == INT{N}_MIN */
+            return f;
+        }
+    }
+
+ overflow:
+    /*
+     * Raise Invalid and return INT{N}_MIN as a float.  Revert any
+     * inexact exception float64_round_to_int may have raised.
+     */
+    set_float_exception_flags(old_flags | float_flag_invalid, fpst);
+    return (uint64_t)(0x800 + 1022 + intsize) << 52;
+}
+
+float64 HELPER(frint32_d)(float64 f, void *fpst)
+{
+    return frint_d(f, fpst, 32);
+}
+
+float64 HELPER(frint64_d)(float64 f, void *fpst)
+{
+    return frint_d(f, fpst, 64);
+}