target-arm: A64: Implement FCVTXN

Implement the FCVTXN operation, which does a narrowing fp precision
conversion using the "round to odd" (von Neumann) mode. This can
conveniently be implemented as "do operation using round to zero;
then set the LSB of the mantissa to 1 if the Inexact flag was set".

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Message-id: 1394822294-14837-24-git-send-email-peter.maydell@linaro.org

target-arm/helper-a64.c

@@ -413,3 +413,26 @@ float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
         return make_float64(sbit | (~exp & 0x7ffULL) << 52);
     }
 }
+
+float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
+{
+    /* Von Neumann rounding is implemented by using round-to-zero
+     * and then setting the LSB of the result if Inexact was raised.
+     */
+    float32 r;
+    float_status *fpst = &env->vfp.fp_status;
+    float_status tstat = *fpst;
+    int exflags;
+
+    set_float_rounding_mode(float_round_to_zero, &tstat);
+    set_float_exception_flags(0, &tstat);
+    r = float64_to_float32(a, &tstat);
+    r = float32_maybe_silence_nan(r);
+    exflags = get_float_exception_flags(&tstat);
+    if (exflags & float_flag_inexact) {
+        r = make_float32(float32_val(r) | 1);
+    }
+    exflags |= get_float_exception_flags(fpst);
+    set_float_exception_flags(exflags, fpst);
+    return r;
+}

target-arm/helper-a64.h

@@ -45,3 +45,4 @@ DEF_HELPER_FLAGS_1(neon_addlp_s16, TCG_CALL_NO_RWG_SE, i64, i64)
 DEF_HELPER_FLAGS_1(neon_addlp_u16, TCG_CALL_NO_RWG_SE, i64, i64)
 DEF_HELPER_FLAGS_2(frecpx_f64, TCG_CALL_NO_RWG, f64, f64, ptr)
 DEF_HELPER_FLAGS_2(frecpx_f32, TCG_CALL_NO_RWG, f32, f32, ptr)
+DEF_HELPER_FLAGS_2(fcvtx_f64_to_f32, TCG_CALL_NO_RWG, f32, f64, env)

target-arm/translate-a64.c

@@ -7278,6 +7278,13 @@ static void handle_2misc_narrow(DisasContext *s, bool scalar,
                 tcg_temp_free_i32(tcg_hi);
             }
             break;
+        case 0x56:  /* FCVTXN, FCVTXN2 */
+            /* 64 bit to 32 bit float conversion
+             * with von Neumann rounding (round to odd)
+             */
+            assert(size == 2);
+            gen_helper_fcvtx_f64_to_f32(tcg_res[pass], tcg_op, cpu_env);
+            break;
         default:
             g_assert_not_reached();
         }
@@ -7391,6 +7398,12 @@ static void disas_simd_scalar_two_reg_misc(DisasContext *s, uint32_t insn)
             rmode = FPROUNDING_TIEAWAY;
             break;
         case 0x56: /* FCVTXN, FCVTXN2 */
+            if (size == 2) {
+                unallocated_encoding(s);
+                return;
+            }
+            handle_2misc_narrow(s, true, opcode, u, false, size - 1, rn, rd);
+            return;
         case 0x7d: /* FRSQRTE */
             unsupported_encoding(s, insn);
             return;
@@ -9244,6 +9257,12 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
         case 0x3d: /* FRECPE */
             handle_2misc_reciprocal(s, opcode, false, u, is_q, size, rn, rd);
             return;
+        case 0x56: /* FCVTXN, FCVTXN2 */
+            if (size == 2) {
+                unallocated_encoding(s);
+                return;
+            }
+            /* fall through */
         case 0x16: /* FCVTN, FCVTN2 */
             /* handle_2misc_narrow does a 2*size -> size operation, but these
              * instructions encode the source size rather than dest size.
@@ -9277,7 +9296,6 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
                 return;
             }
             break;
-        case 0x56: /* FCVTXN, FCVTXN2 */
         case 0x7c: /* URSQRTE */
         case 0x7d: /* FRSQRTE */
             unsupported_encoding(s, insn);