Merge remote-tracking branch 'remotes/rth/tags/pull-sf-20200829' into staging

* float16 comparison wrappers
* float16 conversions to/from 8-bit integers
* bfloat16 support

# gpg: Signature made Sun 30 Aug 2020 03:29:54 BST
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "richard.henderson@linaro.org"
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-sf-20200829:
  softfloat: Define comparison operations for bfloat16
  softfloat: Define misc operations for bfloat16
  softfloat: Define convert operations for bfloat16
  softfloat: Define operations for bfloat16
  softfloat: Add float16_is_normal
  softfloat: Add fp16 and uint8/int8 conversion functions
  softfloat: Implement the full set of comparisons for float16

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

2 files changed, 240 insertions, 0 deletions

diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h
index c7ddcab8ca..8a3f20fae9 100644
--- a/include/fpu/softfloat-types.h
+++ b/include/fpu/softfloat-types.h
@@ -113,6 +113,11 @@ typedef struct {
 #define make_float128_init(high_, low_) { .high = high_, .low = low_ }
 
 /*
+ * Software neural-network floating-point types.
+ */
+typedef uint16_t bfloat16;
+
+/*
  * Software IEC/IEEE floating-point underflow tininess-detection mode.
  */
 
diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h
index 659218b5c7..78ad5ca738 100644
--- a/include/fpu/softfloat.h
+++ b/include/fpu/softfloat.h
@@ -109,6 +109,7 @@ void float_raise(uint8_t flags, float_status *status);
 float16 float16_squash_input_denormal(float16 a, float_status *status);
 float32 float32_squash_input_denormal(float32 a, float_status *status);
 float64 float64_squash_input_denormal(float64 a, float_status *status);
+bfloat16 bfloat16_squash_input_denormal(bfloat16 a, float_status *status);
 
 /*----------------------------------------------------------------------------
 | Options to indicate which negations to perform in float*_muladd()
@@ -136,9 +137,11 @@ float16 uint16_to_float16_scalbn(uint16_t a, int, float_status *status);
 float16 uint32_to_float16_scalbn(uint32_t a, int, float_status *status);
 float16 uint64_to_float16_scalbn(uint64_t a, int, float_status *status);
 
+float16 int8_to_float16(int8_t a, float_status *status);
 float16 int16_to_float16(int16_t a, float_status *status);
 float16 int32_to_float16(int32_t a, float_status *status);
 float16 int64_to_float16(int64_t a, float_status *status);
+float16 uint8_to_float16(uint8_t a, float_status *status);
 float16 uint16_to_float16(uint16_t a, float_status *status);
 float16 uint32_to_float16(uint32_t a, float_status *status);
 float16 uint64_to_float16(uint64_t a, float_status *status);
@@ -187,10 +190,13 @@ float32 float16_to_float32(float16, bool ieee, float_status *status);
 float16 float64_to_float16(float64 a, bool ieee, float_status *status);
 float64 float16_to_float64(float16 a, bool ieee, float_status *status);
 
+int8_t  float16_to_int8_scalbn(float16, FloatRoundMode, int,
+                               float_status *status);
 int16_t float16_to_int16_scalbn(float16, FloatRoundMode, int, float_status *);
 int32_t float16_to_int32_scalbn(float16, FloatRoundMode, int, float_status *);
 int64_t float16_to_int64_scalbn(float16, FloatRoundMode, int, float_status *);
 
+int8_t  float16_to_int8(float16, float_status *status);
 int16_t float16_to_int16(float16, float_status *status);
 int32_t float16_to_int32(float16, float_status *status);
 int64_t float16_to_int64(float16, float_status *status);
@@ -199,6 +205,8 @@ int16_t float16_to_int16_round_to_zero(float16, float_status *status);
 int32_t float16_to_int32_round_to_zero(float16, float_status *status);
 int64_t float16_to_int64_round_to_zero(float16, float_status *status);
 
+uint8_t float16_to_uint8_scalbn(float16 a, FloatRoundMode,
+                                int, float_status *status);
 uint16_t float16_to_uint16_scalbn(float16 a, FloatRoundMode,
                                   int, float_status *status);
 uint32_t float16_to_uint32_scalbn(float16 a, FloatRoundMode,
@@ -206,6 +214,7 @@ uint32_t float16_to_uint32_scalbn(float16 a, FloatRoundMode,
 uint64_t float16_to_uint64_scalbn(float16 a, FloatRoundMode,
                                   int, float_status *status);
 
+uint8_t  float16_to_uint8(float16 a, float_status *status);
 uint16_t float16_to_uint16(float16 a, float_status *status);
 uint32_t float16_to_uint32(float16 a, float_status *status);
 uint64_t float16_to_uint64(float16 a, float_status *status);
@@ -264,6 +273,11 @@ static inline bool float16_is_zero_or_denormal(float16 a)
     return (float16_val(a) & 0x7c00) == 0;
 }
 
+static inline bool float16_is_normal(float16 a)
+{
+    return (((float16_val(a) >> 10) + 1) & 0x1f) >= 2;
+}
+
 static inline float16 float16_abs(float16 a)
 {
     /* Note that abs does *not* handle NaN specially, nor does
@@ -285,6 +299,47 @@ static inline float16 float16_set_sign(float16 a, int sign)
     return make_float16((float16_val(a) & 0x7fff) | (sign << 15));
 }
 
+static inline bool float16_eq(float16 a, float16 b, float_status *s)
+{
+    return float16_compare(a, b, s) == float_relation_equal;
+}
+
+static inline bool float16_le(float16 a, float16 b, float_status *s)
+{
+    return float16_compare(a, b, s) <= float_relation_equal;
+}
+
+static inline bool float16_lt(float16 a, float16 b, float_status *s)
+{
+    return float16_compare(a, b, s) < float_relation_equal;
+}
+
+static inline bool float16_unordered(float16 a, float16 b, float_status *s)
+{
+    return float16_compare(a, b, s) == float_relation_unordered;
+}
+
+static inline bool float16_eq_quiet(float16 a, float16 b, float_status *s)
+{
+    return float16_compare_quiet(a, b, s) == float_relation_equal;
+}
+
+static inline bool float16_le_quiet(float16 a, float16 b, float_status *s)
+{
+    return float16_compare_quiet(a, b, s) <= float_relation_equal;
+}
+
+static inline bool float16_lt_quiet(float16 a, float16 b, float_status *s)
+{
+    return float16_compare_quiet(a, b, s) < float_relation_equal;
+}
+
+static inline bool float16_unordered_quiet(float16 a, float16 b,
+                                           float_status *s)
+{
+    return float16_compare_quiet(a, b, s) == float_relation_unordered;
+}
+
 #define float16_zero make_float16(0)
 #define float16_half make_float16(0x3800)
 #define float16_one make_float16(0x3c00)
@@ -294,6 +349,186 @@ static inline float16 float16_set_sign(float16 a, int sign)
 #define float16_infinity make_float16(0x7c00)
 
 /*----------------------------------------------------------------------------
+| Software bfloat16 conversion routines.
+*----------------------------------------------------------------------------*/
+
+bfloat16 bfloat16_round_to_int(bfloat16, float_status *status);
+bfloat16 float32_to_bfloat16(float32, float_status *status);
+float32 bfloat16_to_float32(bfloat16, float_status *status);
+bfloat16 float64_to_bfloat16(float64 a, float_status *status);
+float64 bfloat16_to_float64(bfloat16 a, float_status *status);
+
+int16_t bfloat16_to_int16_scalbn(bfloat16, FloatRoundMode,
+                                 int, float_status *status);
+int32_t bfloat16_to_int32_scalbn(bfloat16, FloatRoundMode,
+                                 int, float_status *status);
+int64_t bfloat16_to_int64_scalbn(bfloat16, FloatRoundMode,
+                                 int, float_status *status);
+
+int16_t bfloat16_to_int16(bfloat16, float_status *status);
+int32_t bfloat16_to_int32(bfloat16, float_status *status);
+int64_t bfloat16_to_int64(bfloat16, float_status *status);
+
+int16_t bfloat16_to_int16_round_to_zero(bfloat16, float_status *status);
+int32_t bfloat16_to_int32_round_to_zero(bfloat16, float_status *status);
+int64_t bfloat16_to_int64_round_to_zero(bfloat16, float_status *status);
+
+uint16_t bfloat16_to_uint16_scalbn(bfloat16 a, FloatRoundMode,
+                                   int, float_status *status);
+uint32_t bfloat16_to_uint32_scalbn(bfloat16 a, FloatRoundMode,
+                                   int, float_status *status);
+uint64_t bfloat16_to_uint64_scalbn(bfloat16 a, FloatRoundMode,
+                                   int, float_status *status);
+
+uint16_t bfloat16_to_uint16(bfloat16 a, float_status *status);
+uint32_t bfloat16_to_uint32(bfloat16 a, float_status *status);
+uint64_t bfloat16_to_uint64(bfloat16 a, float_status *status);
+
+uint16_t bfloat16_to_uint16_round_to_zero(bfloat16 a, float_status *status);
+uint32_t bfloat16_to_uint32_round_to_zero(bfloat16 a, float_status *status);
+uint64_t bfloat16_to_uint64_round_to_zero(bfloat16 a, float_status *status);
+
+bfloat16 int16_to_bfloat16_scalbn(int16_t a, int, float_status *status);
+bfloat16 int32_to_bfloat16_scalbn(int32_t a, int, float_status *status);
+bfloat16 int64_to_bfloat16_scalbn(int64_t a, int, float_status *status);
+bfloat16 uint16_to_bfloat16_scalbn(uint16_t a, int, float_status *status);
+bfloat16 uint32_to_bfloat16_scalbn(uint32_t a, int, float_status *status);
+bfloat16 uint64_to_bfloat16_scalbn(uint64_t a, int, float_status *status);
+
+bfloat16 int16_to_bfloat16(int16_t a, float_status *status);
+bfloat16 int32_to_bfloat16(int32_t a, float_status *status);
+bfloat16 int64_to_bfloat16(int64_t a, float_status *status);
+bfloat16 uint16_to_bfloat16(uint16_t a, float_status *status);
+bfloat16 uint32_to_bfloat16(uint32_t a, float_status *status);
+bfloat16 uint64_to_bfloat16(uint64_t a, float_status *status);
+
+/*----------------------------------------------------------------------------
+| Software bfloat16 operations.
+*----------------------------------------------------------------------------*/
+
+bfloat16 bfloat16_add(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_sub(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_mul(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_div(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_muladd(bfloat16, bfloat16, bfloat16, int,
+                         float_status *status);
+float16 bfloat16_scalbn(bfloat16, int, float_status *status);
+bfloat16 bfloat16_min(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_max(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_minnum(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_maxnum(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_minnummag(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_maxnummag(bfloat16, bfloat16, float_status *status);
+bfloat16 bfloat16_sqrt(bfloat16, float_status *status);
+FloatRelation bfloat16_compare(bfloat16, bfloat16, float_status *status);
+FloatRelation bfloat16_compare_quiet(bfloat16, bfloat16, float_status *status);
+
+bool bfloat16_is_quiet_nan(bfloat16, float_status *status);
+bool bfloat16_is_signaling_nan(bfloat16, float_status *status);
+bfloat16 bfloat16_silence_nan(bfloat16, float_status *status);
+bfloat16 bfloat16_default_nan(float_status *status);
+
+static inline bool bfloat16_is_any_nan(bfloat16 a)
+{
+    return ((a & ~0x8000) > 0x7F80);
+}
+
+static inline bool bfloat16_is_neg(bfloat16 a)
+{
+    return a >> 15;
+}
+
+static inline bool bfloat16_is_infinity(bfloat16 a)
+{
+    return (a & 0x7fff) == 0x7F80;
+}
+
+static inline bool bfloat16_is_zero(bfloat16 a)
+{
+    return (a & 0x7fff) == 0;
+}
+
+static inline bool bfloat16_is_zero_or_denormal(bfloat16 a)
+{
+    return (a & 0x7F80) == 0;
+}
+
+static inline bool bfloat16_is_normal(bfloat16 a)
+{
+    return (((a >> 7) + 1) & 0xff) >= 2;
+}
+
+static inline bfloat16 bfloat16_abs(bfloat16 a)
+{
+    /* Note that abs does *not* handle NaN specially, nor does
+     * it flush denormal inputs to zero.
+     */
+    return a & 0x7fff;
+}
+
+static inline bfloat16 bfloat16_chs(bfloat16 a)
+{
+    /* Note that chs does *not* handle NaN specially, nor does
+     * it flush denormal inputs to zero.
+     */
+    return a ^ 0x8000;
+}
+
+static inline bfloat16 bfloat16_set_sign(bfloat16 a, int sign)
+{
+    return (a & 0x7fff) | (sign << 15);
+}
+
+static inline bool bfloat16_eq(bfloat16 a, bfloat16 b, float_status *s)
+{
+    return bfloat16_compare(a, b, s) == float_relation_equal;
+}
+
+static inline bool bfloat16_le(bfloat16 a, bfloat16 b, float_status *s)
+{
+    return bfloat16_compare(a, b, s) <= float_relation_equal;
+}
+
+static inline bool bfloat16_lt(bfloat16 a, bfloat16 b, float_status *s)
+{
+    return bfloat16_compare(a, b, s) < float_relation_equal;
+}
+
+static inline bool bfloat16_unordered(bfloat16 a, bfloat16 b, float_status *s)
+{
+    return bfloat16_compare(a, b, s) == float_relation_unordered;
+}
+
+static inline bool bfloat16_eq_quiet(bfloat16 a, bfloat16 b, float_status *s)
+{
+    return bfloat16_compare_quiet(a, b, s) == float_relation_equal;
+}
+
+static inline bool bfloat16_le_quiet(bfloat16 a, bfloat16 b, float_status *s)
+{
+    return bfloat16_compare_quiet(a, b, s) <= float_relation_equal;
+}
+
+static inline bool bfloat16_lt_quiet(bfloat16 a, bfloat16 b, float_status *s)
+{
+    return bfloat16_compare_quiet(a, b, s) < float_relation_equal;
+}
+
+static inline bool bfloat16_unordered_quiet(bfloat16 a, bfloat16 b,
+                                           float_status *s)
+{
+    return bfloat16_compare_quiet(a, b, s) == float_relation_unordered;
+}
+
+#define bfloat16_zero 0
+#define bfloat16_half 0x3f00
+#define bfloat16_one 0x3f80
+#define bfloat16_one_point_five 0x3fc0
+#define bfloat16_two 0x4000
+#define bfloat16_three 0x4040
+#define bfloat16_infinity 0x7f80
+
+/*----------------------------------------------------------------------------
 | The pattern for a default generated half-precision NaN.
 *----------------------------------------------------------------------------*/
 float16 float16_default_nan(float_status *status);