softfloat: Allow runtime choice of inf * 0 + NaN result

IEEE 758 does not define a fixed rule for what NaN to return in
the case of a fused multiply-add of inf * 0 + NaN. Different
architectures thus do different things:
 * some return the default NaN
 * some return the input NaN
 * Arm returns the default NaN if the input NaN is quiet,
   and the input NaN if it is signalling

We want to make this logic be runtime selected rather than
hardcoded into the binary, because:
 * this will let us have multiple targets in one QEMU binary
 * the Arm FEAT_AFP architectural feature includes letting
   the guest select a NaN propagation rule at runtime

In this commit we add an enum for the propagation rule, the field in
float_status, and the corresponding getters and setters.  We change
pickNaNMulAdd to honour this, but because all targets still leave
this field at its default 0 value, the fallback logic will pick the
rule type with the old ifdef ladder.

Note that four architectures both use the muladd softfloat functions
and did not have a branch of the ifdef ladder to specify their
behaviour (and so were ending up with the "default" case, probably
wrongly): i386, HPPA, SH4 and Tricore.  SH4 and Tricore both set
default_nan_mode, and so will never get into pickNaNMulAdd().  For
HPPA and i386 we retain the same behaviour as the old default-case,
which is to not ever return the default NaN.  This might not be
correct but it is not a behaviour change.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20241202131347.498124-4-peter.maydell@linaro.org

diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc
index 81a67eb67b52dbdfc0e91bc316e15e0445f7f0f9..f5b422e07b51e34bf97be472451d0e31915b2647 100644 (file)
--- a/fpu/softfloat-specialize.c.inc
+++ b/fpu/softfloat-specialize.c.inc
@@ -475,6 +475,8 @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls,
 static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
                          bool infzero, float_status *status)
 {
+    FloatInfZeroNaNRule rule = status->float_infzeronan_rule;
+
     /*
      * We guarantee not to require the target to tell us how to
      * pick a NaN if we're always returning the default NaN.
@@ -482,14 +484,68 @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
      * specify.
      */
     assert(!status->default_nan_mode);
+
+    if (rule == float_infzeronan_none) {
+        /*
+         * Temporarily fall back to ifdef ladder
+         */
 #if defined(TARGET_ARM)
-    /* For ARM, the (inf,zero,qnan) case sets InvalidOp and returns
-     * the default NaN
-     */
-    if (infzero && is_qnan(c_cls)) {
-        return 3;
+        /*
+         * For ARM, the (inf,zero,qnan) case returns the default NaN,
+         * but (inf,zero,snan) returns the input NaN.
+         */
+        rule = float_infzeronan_dnan_if_qnan;
+#elif defined(TARGET_MIPS)
+        if (snan_bit_is_one(status)) {
+            /*
+             * For MIPS systems that conform to IEEE754-1985, the (inf,zero,nan)
+             * case sets InvalidOp and returns the default NaN
+             */
+            rule = float_infzeronan_dnan_always;
+        } else {
+            /*
+             * For MIPS systems that conform to IEEE754-2008, the (inf,zero,nan)
+             * case sets InvalidOp and returns the input value 'c'
+             */
+            rule = float_infzeronan_dnan_never;
+        }
+#elif defined(TARGET_PPC) || defined(TARGET_SPARC) || \
+    defined(TARGET_XTENSA) || defined(TARGET_HPPA) || \
+    defined(TARGET_I386) || defined(TARGET_LOONGARCH)
+        /*
+         * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan)
+         * case sets InvalidOp and returns the input value 'c'
+         */
+        /*
+         * For PPC, the (inf,zero,qnan) case sets InvalidOp, but we prefer
+         * to return an input NaN if we have one (ie c) rather than generating
+         * a default NaN
+         */
+        rule = float_infzeronan_dnan_never;
+#elif defined(TARGET_S390X)
+        rule = float_infzeronan_dnan_always;
+#endif
     }
 
+    if (infzero) {
+        /*
+         * Inf * 0 + NaN -- some implementations return the default NaN here,
+         * and some return the input NaN.
+         */
+        switch (rule) {
+        case float_infzeronan_dnan_never:
+            return 2;
+        case float_infzeronan_dnan_always:
+            return 3;
+        case float_infzeronan_dnan_if_qnan:
+            return is_qnan(c_cls) ? 3 : 2;
+        default:
+            g_assert_not_reached();
+        }
+    }
+
+#if defined(TARGET_ARM)
+
     /* This looks different from the ARM ARM pseudocode, because the ARM ARM
      * puts the operands to a fused mac operation (a*b)+c in the order c,a,b.
      */
@@ -508,13 +564,6 @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
     }
 #elif defined(TARGET_MIPS)
     if (snan_bit_is_one(status)) {
-        /*
-         * For MIPS systems that conform to IEEE754-1985, the (inf,zero,nan)
-         * case sets InvalidOp and returns the default NaN
-         */
-        if (infzero) {
-            return 3;
-        }
         /* Prefer sNaN over qNaN, in the a, b, c order. */
         if (is_snan(a_cls)) {
             return 0;
@@ -530,10 +579,6 @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
             return 2;
         }
     } else {
-        /*
-         * For MIPS systems that conform to IEEE754-2008, the (inf,zero,nan)
-         * case sets InvalidOp and returns the input value 'c'
-         */
         /* Prefer sNaN over qNaN, in the c, a, b order. */
         if (is_snan(c_cls)) {
             return 2;
@@ -550,11 +595,6 @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
         }
     }
 #elif defined(TARGET_LOONGARCH64)
-    /*
-     * For LoongArch systems that conform to IEEE754-2008, the (inf,zero,nan)
-     * case sets InvalidOp and returns the input value 'c'
-     */
-
     /* Prefer sNaN over qNaN, in the c, a, b order. */
     if (is_snan(c_cls)) {
         return 2;
@@ -570,11 +610,6 @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
         return 1;
     }
 #elif defined(TARGET_PPC)
-    /* For PPC, the (inf,zero,qnan) case sets InvalidOp, but we prefer
-     * to return an input NaN if we have one (ie c) rather than generating
-     * a default NaN
-     */
-
     /* If fRA is a NaN return it; otherwise if fRB is a NaN return it;
      * otherwise return fRC. Note that muladd on PPC is (fRA * fRC) + frB
      */
@@ -586,10 +621,6 @@ static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
         return 1;
     }
 #elif defined(TARGET_S390X)
-    if (infzero) {
-        return 3;
-    }
-
     if (is_snan(a_cls)) {
         return 0;
     } else if (is_snan(b_cls)) {

diff --git a/include/fpu/softfloat-helpers.h b/include/fpu/softfloat-helpers.h
index 453188de70b3e761e6bed15f64cb75fbbf851fc9..0bf44dc608722f11ed03575f705d9fb029c845b6 100644 (file)
--- a/include/fpu/softfloat-helpers.h
+++ b/include/fpu/softfloat-helpers.h
@@ -81,6 +81,12 @@ static inline void set_float_2nan_prop_rule(Float2NaNPropRule rule,
     status->float_2nan_prop_rule = rule;
 }
 
+static inline void set_float_infzeronan_rule(FloatInfZeroNaNRule rule,
+                                             float_status *status)
+{
+    status->float_infzeronan_rule = rule;
+}
+
 static inline void set_flush_to_zero(bool val, float_status *status)
 {
     status->flush_to_zero = val;
@@ -137,6 +143,11 @@ static inline Float2NaNPropRule get_float_2nan_prop_rule(float_status *status)
     return status->float_2nan_prop_rule;
 }
 
+static inline FloatInfZeroNaNRule get_float_infzeronan_rule(float_status *status)
+{
+    return status->float_infzeronan_rule;
+}
+
 static inline bool get_flush_to_zero(float_status *status)
 {
     return status->flush_to_zero;

diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h
index 8f39691dfd0c2d6968e8dbe40eb3c2d3c068ffee..47bb22c4e254db1ad1deb0c1420afaf994c2d0ce 100644 (file)
--- a/include/fpu/softfloat-types.h
+++ b/include/fpu/softfloat-types.h
@@ -207,6 +207,28 @@ typedef enum __attribute__((__packed__)) {
     float_2nan_prop_x87,
 } Float2NaNPropRule;
 
+/*
+ * Rule for result of fused multiply-add 0 * Inf + NaN.
+ * This must be a NaN, but implementations differ on whether this
+ * is the input NaN or the default NaN.
+ *
+ * You don't need to set this if default_nan_mode is enabled.
+ * When not in default-NaN mode, it is an error for the target
+ * not to set the rule in float_status if it uses muladd, and we
+ * will assert if we need to handle an input NaN and no rule was
+ * selected.
+ */
+typedef enum __attribute__((__packed__)) {
+    /* No propagation rule specified */
+    float_infzeronan_none = 0,
+    /* Result is never the default NaN (so always the input NaN) */
+    float_infzeronan_dnan_never,
+    /* Result is always the default NaN */
+    float_infzeronan_dnan_always,
+    /* Result is the default NaN if the input NaN is quiet */
+    float_infzeronan_dnan_if_qnan,
+} FloatInfZeroNaNRule;
+
 /*
  * Floating Point Status. Individual architectures may maintain
  * several versions of float_status for different functions. The
@@ -219,6 +241,7 @@ typedef struct float_status {
     FloatRoundMode float_rounding_mode;
     FloatX80RoundPrec floatx80_rounding_precision;
     Float2NaNPropRule float_2nan_prop_rule;
+    FloatInfZeroNaNRule float_infzeronan_rule;
     bool tininess_before_rounding;
     /* should denormalised results go to zero and set the inexact flag? */
     bool flush_to_zero;