[Qemu-devel] [PATCH v5 00/28] softfloat patch roundup

[Qemu-devel] [PATCH v5 00/28] softfloat patch roundup

[Richard Henderson]

This is my SNaN patch set, Alex's float-float refactor, and a couple
of other random outstanding fpu patches.  This has been reordered so
as to be bisectable, since the float-float refactor requires the snan
work to avoid breakage.

This was built on top of pm215/target-arm.next to make it easier to test.
The whole tree is

  git://github.com/rth7680/qemu.git fpu-roundup

Changes since last:
  - Use v5 instead of v3, since Alex was up to v4 with float-float.
  - Incorporate feedback from pm215.
  - Include float128 fix from Petr.
  - Include further cleanups for default_nan and silence_nan.
    The first of these was inspired by a comment from pm215; 
    the rest follow as logical extensions.


r~


Alex Bennée (4):
  target/arm: convert conversion helpers to fpst/ahp_flag
  target/arm: squash FZ16 behaviour for conversions
  fpu/softfloat: Partial support for ARM Alternative half-precision
  fpu/softfloat: re-factor float to float conversions

Petr Tesarik (1):
  fpu/softfloat: Fix conversion from uint64 to float128

Richard Henderson (23):
  fpu/softfloat: Merge NO_SIGNALING_NANS definitions
  fpu/softfloat: Split floatXX_silence_nan from floatXX_maybe_silence_nan
  fpu/softfloat: Move softfloat-specialize.h below FloatParts definition
  fpu/softfloat: Canonicalize NaN fraction
  fpu/softfloat: Introduce parts_is_snan_frac
  fpu/softfloat: Replace float_class_dnan with parts_default_nan
  fpu/softfloat: Replace float_class_msnan with parts_silence_nan
  target/arm: Use floatX_silence_nan when we have already checked for SNaN
  target/arm: Remove floatX_maybe_silence_nan from conversions
  target/hppa: Remove floatX_maybe_silence_nan from conversions
  target/m68k: Use floatX_silence_nan when we have already checked for SNaN
  target/mips: Remove floatX_maybe_silence_nan from conversions
  target/riscv: Remove floatX_maybe_silence_nan from conversions
  target/s390x: Remove floatX_maybe_silence_nan from conversions
  fpu/softfloat: Use float*_silence_nan in propagateFloat*NaN
  fpu/softfloat: Remove floatX_maybe_silence_nan
  fpu/softfloat: Specialize on snan_bit_is_one
  fpu/softfloat: Make is_nan et al available to softfloat-specialize.h
  fpu/softfloat: Pass FloatClass to pickNaN
  fpu/softfloat: Pass FloatClass to pickNaNMulAdd
  fpu/softfloat: Define floatN_default_nan in terms of parts_default_nan
  fpu/softfloat: Clean up parts_default_nan
  fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan

 fpu/softfloat-specialize.h    | 691 ++++++++++++++-------------------
 include/fpu/softfloat-types.h |   1 +
 include/fpu/softfloat.h       |  18 +-
 target/arm/helper.h           |  10 +-
 target/arm/translate.h        |  12 +
 fpu/softfloat.c               | 708 ++++++++++++----------------------
 target/arm/helper-a64.c       |   7 +-
 target/arm/helper.c           | 114 +++---
 target/arm/translate-a64.c    |  38 +-
 target/arm/translate.c        |  74 +++-
 target/hppa/cpu.c             |   1 -
 target/hppa/op_helper.c       |   2 -
 target/m68k/softfloat.c       |   3 +-
 target/mips/msa_helper.c      |   4 -
 target/mips/op_helper.c       |   2 -
 target/ppc/fpu_helper.c       |   1 -
 target/riscv/fpu_helper.c     |   6 +-
 target/s390x/fpu_helper.c     |  12 +-
 target/sh4/cpu.c              |   1 -
 target/unicore32/cpu.c        |   2 -
 20 files changed, 701 insertions(+), 1006 deletions(-)

-- 
2.17.0

[Qemu-devel] [PATCH v5 01/28] fpu/softfloat: Fix conversion from uint64 to float128

[Richard Henderson]

From: Petr Tesarik <ptesarik@suse.com>

The significand is passed to normalizeRoundAndPackFloat128() as high
first, low second. The current code passes the integer first, so the
result is incorrectly shifted left by 64 bits.

This bug affects the emulation of s390x instruction CXLGBR (convert
from logical 64-bit binary-integer operand to extended BFP result).

Cc: qemu-stable@nongnu.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Petr Tesarik <ptesarik@suse.com>
Message-Id: <20180511071052.1443-1-ptesarik@suse.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index bc0f52fa54..d07419324a 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -3147,7 +3147,7 @@ float128 uint64_to_float128(uint64_t a, float_status *status)
     if (a == 0) {
         return float128_zero;
     }
-    return normalizeRoundAndPackFloat128(0, 0x406E, a, 0, status);
+    return normalizeRoundAndPackFloat128(0, 0x406E, 0, a, status);
 }
 
 
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 01/28] fpu/softfloat: Fix conversion from uint64 to float128

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> From: Petr Tesarik <ptesarik@suse.com>
>
> The significand is passed to normalizeRoundAndPackFloat128() as high
> first, low second. The current code passes the integer first, so the
> result is incorrectly shifted left by 64 bits.
>
> This bug affects the emulation of s390x instruction CXLGBR (convert
> from logical 64-bit binary-integer operand to extended BFP result).
>
> Cc: qemu-stable@nongnu.org
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Petr Tesarik <ptesarik@suse.com>
> Message-Id: <20180511071052.1443-1-ptesarik@suse.com>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  fpu/softfloat.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index bc0f52fa54..d07419324a 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -3147,7 +3147,7 @@ float128 uint64_to_float128(uint64_t a, float_status *status)
>      if (a == 0) {
>          return float128_zero;
>      }
> -    return normalizeRoundAndPackFloat128(0, 0x406E, a, 0, status);
> +    return normalizeRoundAndPackFloat128(0, 0x406E, 0, a, status);

This brings back memories :-/

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

--
Alex Bennée

[Qemu-devel] [PATCH v5 02/28] fpu/softfloat: Merge NO_SIGNALING_NANS definitions

[Richard Henderson]

Move the ifdef inside the relevant functions instead of
duplicating the function declarations.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 100 +++++++++++++++----------------------
 1 file changed, 40 insertions(+), 60 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 27834af0de..58b05718c8 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -233,17 +233,6 @@ typedef struct {
     uint64_t high, low;
 } commonNaNT;
 
-#ifdef NO_SIGNALING_NANS
-int float16_is_quiet_nan(float16 a_, float_status *status)
-{
-    return float16_is_any_nan(a_);
-}
-
-int float16_is_signaling_nan(float16 a_, float_status *status)
-{
-    return 0;
-}
-#else
 /*----------------------------------------------------------------------------
 | Returns 1 if the half-precision floating-point value `a' is a quiet
 | NaN; otherwise returns 0.
@@ -251,12 +240,16 @@ int float16_is_signaling_nan(float16 a_, float_status *status)
 
 int float16_is_quiet_nan(float16 a_, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return float16_is_any_nan(a_);
+#else
     uint16_t a = float16_val(a_);
     if (status->snan_bit_is_one) {
         return (((a >> 9) & 0x3F) == 0x3E) && (a & 0x1FF);
     } else {
         return ((a & ~0x8000) >= 0x7C80);
     }
+#endif
 }
 
 /*----------------------------------------------------------------------------
@@ -266,14 +259,17 @@ int float16_is_quiet_nan(float16 a_, float_status *status)
 
 int float16_is_signaling_nan(float16 a_, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return 0;
+#else
     uint16_t a = float16_val(a_);
     if (status->snan_bit_is_one) {
         return ((a & ~0x8000) >= 0x7C80);
     } else {
         return (((a >> 9) & 0x3F) == 0x3E) && (a & 0x1FF);
     }
-}
 #endif
+}
 
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the half-precision floating point value `a' is a
@@ -333,17 +329,6 @@ static float16 commonNaNToFloat16(commonNaNT a, float_status *status)
     }
 }
 
-#ifdef NO_SIGNALING_NANS
-int float32_is_quiet_nan(float32 a_, float_status *status)
-{
-    return float32_is_any_nan(a_);
-}
-
-int float32_is_signaling_nan(float32 a_, float_status *status)
-{
-    return 0;
-}
-#else
 /*----------------------------------------------------------------------------
 | Returns 1 if the single-precision floating-point value `a' is a quiet
 | NaN; otherwise returns 0.
@@ -351,12 +336,16 @@ int float32_is_signaling_nan(float32 a_, float_status *status)
 
 int float32_is_quiet_nan(float32 a_, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return float32_is_any_nan(a_);
+#else
     uint32_t a = float32_val(a_);
     if (status->snan_bit_is_one) {
         return (((a >> 22) & 0x1FF) == 0x1FE) && (a & 0x003FFFFF);
     } else {
         return ((uint32_t)(a << 1) >= 0xFF800000);
     }
+#endif
 }
 
 /*----------------------------------------------------------------------------
@@ -366,14 +355,17 @@ int float32_is_quiet_nan(float32 a_, float_status *status)
 
 int float32_is_signaling_nan(float32 a_, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return 0;
+#else
     uint32_t a = float32_val(a_);
     if (status->snan_bit_is_one) {
         return ((uint32_t)(a << 1) >= 0xFF800000);
     } else {
         return (((a >> 22) & 0x1FF) == 0x1FE) && (a & 0x003FFFFF);
     }
-}
 #endif
+}
 
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the single-precision floating point value `a' is a
@@ -744,17 +736,6 @@ static float32 propagateFloat32NaN(float32 a, float32 b, float_status *status)
     }
 }
 
-#ifdef NO_SIGNALING_NANS
-int float64_is_quiet_nan(float64 a_, float_status *status)
-{
-    return float64_is_any_nan(a_);
-}
-
-int float64_is_signaling_nan(float64 a_, float_status *status)
-{
-    return 0;
-}
-#else
 /*----------------------------------------------------------------------------
 | Returns 1 if the double-precision floating-point value `a' is a quiet
 | NaN; otherwise returns 0.
@@ -762,6 +743,9 @@ int float64_is_signaling_nan(float64 a_, float_status *status)
 
 int float64_is_quiet_nan(float64 a_, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return float64_is_any_nan(a_);
+#else
     uint64_t a = float64_val(a_);
     if (status->snan_bit_is_one) {
         return (((a >> 51) & 0xFFF) == 0xFFE)
@@ -769,6 +753,7 @@ int float64_is_quiet_nan(float64 a_, float_status *status)
     } else {
         return ((a << 1) >= 0xFFF0000000000000ULL);
     }
+#endif
 }
 
 /*----------------------------------------------------------------------------
@@ -778,6 +763,9 @@ int float64_is_quiet_nan(float64 a_, float_status *status)
 
 int float64_is_signaling_nan(float64 a_, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return 0;
+#else
     uint64_t a = float64_val(a_);
     if (status->snan_bit_is_one) {
         return ((a << 1) >= 0xFFF0000000000000ULL);
@@ -785,8 +773,8 @@ int float64_is_signaling_nan(float64 a_, float_status *status)
         return (((a >> 51) & 0xFFF) == 0xFFE)
             && (a & LIT64(0x0007FFFFFFFFFFFF));
     }
-}
 #endif
+}
 
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the double-precision floating point value `a' is a
@@ -899,17 +887,6 @@ static float64 propagateFloat64NaN(float64 a, float64 b, float_status *status)
     }
 }
 
-#ifdef NO_SIGNALING_NANS
-int floatx80_is_quiet_nan(floatx80 a_, float_status *status)
-{
-    return floatx80_is_any_nan(a_);
-}
-
-int floatx80_is_signaling_nan(floatx80 a_, float_status *status)
-{
-    return 0;
-}
-#else
 /*----------------------------------------------------------------------------
 | Returns 1 if the extended double-precision floating-point value `a' is a
 | quiet NaN; otherwise returns 0. This slightly differs from the same
@@ -918,6 +895,9 @@ int floatx80_is_signaling_nan(floatx80 a_, float_status *status)
 
 int floatx80_is_quiet_nan(floatx80 a, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return floatx80_is_any_nan(a);
+#else
     if (status->snan_bit_is_one) {
         uint64_t aLow;
 
@@ -929,6 +909,7 @@ int floatx80_is_quiet_nan(floatx80 a, float_status *status)
         return ((a.high & 0x7FFF) == 0x7FFF)
             && (LIT64(0x8000000000000000) <= ((uint64_t)(a.low << 1)));
     }
+#endif
 }
 
 /*----------------------------------------------------------------------------
@@ -939,6 +920,9 @@ int floatx80_is_quiet_nan(floatx80 a, float_status *status)
 
 int floatx80_is_signaling_nan(floatx80 a, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return 0;
+#else
     if (status->snan_bit_is_one) {
         return ((a.high & 0x7FFF) == 0x7FFF)
             && ((a.low << 1) >= 0x8000000000000000ULL);
@@ -950,8 +934,8 @@ int floatx80_is_signaling_nan(floatx80 a, float_status *status)
             && (uint64_t)(aLow << 1)
             && (a.low == aLow);
     }
-}
 #endif
+}
 
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the extended double-precision floating point value
@@ -1060,17 +1044,6 @@ floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status)
     }
 }
 
-#ifdef NO_SIGNALING_NANS
-int float128_is_quiet_nan(float128 a_, float_status *status)
-{
-    return float128_is_any_nan(a_);
-}
-
-int float128_is_signaling_nan(float128 a_, float_status *status)
-{
-    return 0;
-}
-#else
 /*----------------------------------------------------------------------------
 | Returns 1 if the quadruple-precision floating-point value `a' is a quiet
 | NaN; otherwise returns 0.
@@ -1078,6 +1051,9 @@ int float128_is_signaling_nan(float128 a_, float_status *status)
 
 int float128_is_quiet_nan(float128 a, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return float128_is_any_nan(a);
+#else
     if (status->snan_bit_is_one) {
         return (((a.high >> 47) & 0xFFFF) == 0xFFFE)
             && (a.low || (a.high & 0x00007FFFFFFFFFFFULL));
@@ -1085,6 +1061,7 @@ int float128_is_quiet_nan(float128 a, float_status *status)
         return ((a.high << 1) >= 0xFFFF000000000000ULL)
             && (a.low || (a.high & 0x0000FFFFFFFFFFFFULL));
     }
+#endif
 }
 
 /*----------------------------------------------------------------------------
@@ -1094,6 +1071,9 @@ int float128_is_quiet_nan(float128 a, float_status *status)
 
 int float128_is_signaling_nan(float128 a, float_status *status)
 {
+#ifdef NO_SIGNALING_NANS
+    return 0;
+#else
     if (status->snan_bit_is_one) {
         return ((a.high << 1) >= 0xFFFF000000000000ULL)
             && (a.low || (a.high & 0x0000FFFFFFFFFFFFULL));
@@ -1101,8 +1081,8 @@ int float128_is_signaling_nan(float128 a, float_status *status)
         return (((a.high >> 47) & 0xFFFF) == 0xFFFE)
             && (a.low || (a.high & LIT64(0x00007FFFFFFFFFFF)));
     }
-}
 #endif
+}
 
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the quadruple-precision floating point value `a' is
-- 
2.17.0

[Qemu-devel] [PATCH v5 03/28] fpu/softfloat: Split floatXX_silence_nan from floatXX_maybe_silence_nan

[Richard Henderson]

The new function assumes that the input is an SNaN and
does not double-check.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 174 +++++++++++++++++++++++++------------
 include/fpu/softfloat.h    |   5 ++
 2 files changed, 123 insertions(+), 56 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 58b05718c8..4fc9ea4ac0 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -271,22 +271,35 @@ int float16_is_signaling_nan(float16 a_, float_status *status)
 #endif
 }
 
+/*----------------------------------------------------------------------------
+| Returns a quiet NaN from a signalling NaN for the half-precision
+| floating point value `a'.
+*----------------------------------------------------------------------------*/
+
+float16 float16_silence_nan(float16 a, float_status *status)
+{
+#ifdef NO_SIGNALING_NANS
+    g_assert_not_reached();
+#else
+    if (status->snan_bit_is_one) {
+        return float16_default_nan(status);
+    } else {
+        return a | (1 << 9);
+    }
+#endif
+}
+
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the half-precision floating point value `a' is a
 | signaling NaN; otherwise returns `a'.
 *----------------------------------------------------------------------------*/
-float16 float16_maybe_silence_nan(float16 a_, float_status *status)
+
+float16 float16_maybe_silence_nan(float16 a, float_status *status)
 {
-    if (float16_is_signaling_nan(a_, status)) {
-        if (status->snan_bit_is_one) {
-            return float16_default_nan(status);
-        } else {
-            uint16_t a = float16_val(a_);
-            a |= (1 << 9);
-            return make_float16(a);
-        }
+    if (float16_is_signaling_nan(a, status)) {
+        return float16_silence_nan(a, status);
     }
-    return a_;
+    return a;
 }
 
 /*----------------------------------------------------------------------------
@@ -367,30 +380,40 @@ int float32_is_signaling_nan(float32 a_, float_status *status)
 #endif
 }
 
+/*----------------------------------------------------------------------------
+| Returns a quiet NaN from a signalling NaN for the single-precision
+| floating point value `a'.
+*----------------------------------------------------------------------------*/
+
+float32 float32_silence_nan(float32 a, float_status *status)
+{
+#ifdef NO_SIGNALING_NANS
+    g_assert_not_reached();
+#else
+    if (status->snan_bit_is_one) {
+# ifdef TARGET_HPPA
+        a &= ~0x00400000;
+        a |=  0x00200000;
+        return a;
+# else
+        return float32_default_nan(status);
+# endif
+    } else {
+        return a | (1 << 22);
+    }
+#endif
+}
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the single-precision floating point value `a' is a
 | signaling NaN; otherwise returns `a'.
 *----------------------------------------------------------------------------*/
 
-float32 float32_maybe_silence_nan(float32 a_, float_status *status)
+float32 float32_maybe_silence_nan(float32 a, float_status *status)
 {
-    if (float32_is_signaling_nan(a_, status)) {
-        if (status->snan_bit_is_one) {
-#ifdef TARGET_HPPA
-            uint32_t a = float32_val(a_);
-            a &= ~0x00400000;
-            a |=  0x00200000;
-            return make_float32(a);
-#else
-            return float32_default_nan(status);
-#endif
-        } else {
-            uint32_t a = float32_val(a_);
-            a |= (1 << 22);
-            return make_float32(a);
-        }
+    if (float32_is_signaling_nan(a, status)) {
+        return float32_silence_nan(a, status);
     }
-    return a_;
+    return a;
 }
 
 /*----------------------------------------------------------------------------
@@ -776,30 +799,41 @@ int float64_is_signaling_nan(float64 a_, float_status *status)
 #endif
 }
 
+/*----------------------------------------------------------------------------
+| Returns a quiet NaN from a signalling NaN for the double-precision
+| floating point value `a'.
+*----------------------------------------------------------------------------*/
+
+float64 float64_silence_nan(float64 a, float_status *status)
+{
+#ifdef NO_SIGNALING_NANS
+    g_assert_not_reached();
+#else
+    if (status->snan_bit_is_one) {
+# ifdef TARGET_HPPA
+        a &= ~0x0008000000000000ULL;
+        a |=  0x0004000000000000ULL;
+        return a;
+# else
+        return float64_default_nan(status);
+# endif
+    } else {
+        return a | LIT64(0x0008000000000000);
+    }
+#endif
+}
+
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the double-precision floating point value `a' is a
 | signaling NaN; otherwise returns `a'.
 *----------------------------------------------------------------------------*/
 
-float64 float64_maybe_silence_nan(float64 a_, float_status *status)
+float64 float64_maybe_silence_nan(float64 a, float_status *status)
 {
-    if (float64_is_signaling_nan(a_, status)) {
-        if (status->snan_bit_is_one) {
-#ifdef TARGET_HPPA
-            uint64_t a = float64_val(a_);
-            a &= ~0x0008000000000000ULL;
-            a |=  0x0004000000000000ULL;
-            return make_float64(a);
-#else
-            return float64_default_nan(status);
-#endif
-        } else {
-            uint64_t a = float64_val(a_);
-            a |= LIT64(0x0008000000000000);
-            return make_float64(a);
-        }
+    if (float64_is_signaling_nan(a, status)) {
+        return float64_silence_nan(a, status);
     }
-    return a_;
+    return a;
 }
 
 /*----------------------------------------------------------------------------
@@ -937,6 +971,25 @@ int floatx80_is_signaling_nan(floatx80 a, float_status *status)
 #endif
 }
 
+/*----------------------------------------------------------------------------
+| Returns a quiet NaN from a signalling NaN for the extended double-precision
+| floating point value `a'.
+*----------------------------------------------------------------------------*/
+
+floatx80 floatx80_silence_nan(floatx80 a, float_status *status)
+{
+#ifdef NO_SIGNALING_NANS
+    g_assert_not_reached();
+#else
+    if (status->snan_bit_is_one) {
+        return floatx80_default_nan(status);
+    } else {
+        a.low |= LIT64(0xC000000000000000);
+        return a;
+    }
+#endif
+}
+
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the extended double-precision floating point value
 | `a' is a signaling NaN; otherwise returns `a'.
@@ -945,12 +998,7 @@ int floatx80_is_signaling_nan(floatx80 a, float_status *status)
 floatx80 floatx80_maybe_silence_nan(floatx80 a, float_status *status)
 {
     if (floatx80_is_signaling_nan(a, status)) {
-        if (status->snan_bit_is_one) {
-            a = floatx80_default_nan(status);
-        } else {
-            a.low |= LIT64(0xC000000000000000);
-            return a;
-        }
+        return floatx80_silence_nan(a, status);
     }
     return a;
 }
@@ -1084,6 +1132,25 @@ int float128_is_signaling_nan(float128 a, float_status *status)
 #endif
 }
 
+/*----------------------------------------------------------------------------
+| Returns a quiet NaN from a signalling NaN for the quadruple-precision
+| floating point value `a'.
+*----------------------------------------------------------------------------*/
+
+float128 float128_silence_nan(float128 a, float_status *status)
+{
+#ifdef NO_SIGNALING_NANS
+    g_assert_not_reached();
+#else
+    if (status->snan_bit_is_one) {
+        return float128_default_nan(status);
+    } else {
+        a.high |= LIT64(0x0000800000000000);
+        return a;
+    }
+#endif
+}
+
 /*----------------------------------------------------------------------------
 | Returns a quiet NaN if the quadruple-precision floating point value `a' is
 | a signaling NaN; otherwise returns `a'.
@@ -1092,12 +1159,7 @@ int float128_is_signaling_nan(float128 a, float_status *status)
 float128 float128_maybe_silence_nan(float128 a, float_status *status)
 {
     if (float128_is_signaling_nan(a, status)) {
-        if (status->snan_bit_is_one) {
-            a = float128_default_nan(status);
-        } else {
-            a.high |= LIT64(0x0000800000000000);
-            return a;
-        }
+        return float128_silence_nan(a, status);
     }
     return a;
 }
diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h
index 36626a501b..43962dc3f5 100644
--- a/include/fpu/softfloat.h
+++ b/include/fpu/softfloat.h
@@ -257,6 +257,7 @@ int float16_compare_quiet(float16, float16, float_status *status);
 
 int float16_is_quiet_nan(float16, float_status *status);
 int float16_is_signaling_nan(float16, float_status *status);
+float16 float16_silence_nan(float16, float_status *status);
 float16 float16_maybe_silence_nan(float16, float_status *status);
 
 static inline int float16_is_any_nan(float16 a)
@@ -368,6 +369,7 @@ float32 float32_minnummag(float32, float32, float_status *status);
 float32 float32_maxnummag(float32, float32, float_status *status);
 int float32_is_quiet_nan(float32, float_status *status);
 int float32_is_signaling_nan(float32, float_status *status);
+float32 float32_silence_nan(float32, float_status *status);
 float32 float32_maybe_silence_nan(float32, float_status *status);
 float32 float32_scalbn(float32, int, float_status *status);
 
@@ -497,6 +499,7 @@ float64 float64_minnummag(float64, float64, float_status *status);
 float64 float64_maxnummag(float64, float64, float_status *status);
 int float64_is_quiet_nan(float64 a, float_status *status);
 int float64_is_signaling_nan(float64, float_status *status);
+float64 float64_silence_nan(float64, float_status *status);
 float64 float64_maybe_silence_nan(float64, float_status *status);
 float64 float64_scalbn(float64, int, float_status *status);
 
@@ -600,6 +603,7 @@ int floatx80_compare(floatx80, floatx80, float_status *status);
 int floatx80_compare_quiet(floatx80, floatx80, float_status *status);
 int floatx80_is_quiet_nan(floatx80, float_status *status);
 int floatx80_is_signaling_nan(floatx80, float_status *status);
+floatx80 floatx80_silence_nan(floatx80, float_status *status);
 floatx80 floatx80_maybe_silence_nan(floatx80, float_status *status);
 floatx80 floatx80_scalbn(floatx80, int, float_status *status);
 
@@ -811,6 +815,7 @@ int float128_compare(float128, float128, float_status *status);
 int float128_compare_quiet(float128, float128, float_status *status);
 int float128_is_quiet_nan(float128, float_status *status);
 int float128_is_signaling_nan(float128, float_status *status);
+float128 float128_silence_nan(float128, float_status *status);
 float128 float128_maybe_silence_nan(float128, float_status *status);
 float128 float128_scalbn(float128, int, float_status *status);
 
-- 
2.17.0

[Qemu-devel] [PATCH v5 04/28] fpu/softfloat: Move softfloat-specialize.h below FloatParts definition

[Richard Henderson]

We want to be able to specialize on the canonical representation.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index d07419324a..0d17027379 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -95,16 +95,6 @@ this code that are retained.
 *----------------------------------------------------------------------------*/
 #include "fpu/softfloat-macros.h"
 
-/*----------------------------------------------------------------------------
-| Functions and definitions to determine:  (1) whether tininess for underflow
-| is detected before or after rounding by default, (2) what (if anything)
-| happens when exceptions are raised, (3) how signaling NaNs are distinguished
-| from quiet NaNs, (4) the default generated quiet NaNs, and (5) how NaNs
-| are propagated from function inputs to output.  These details are target-
-| specific.
-*----------------------------------------------------------------------------*/
-#include "softfloat-specialize.h"
-
 /*----------------------------------------------------------------------------
 | Returns the fraction bits of the half-precision floating-point value `a'.
 *----------------------------------------------------------------------------*/
@@ -322,6 +312,16 @@ static inline float64 float64_pack_raw(FloatParts p)
     return make_float64(pack_raw(float64_params, p));
 }
 
+/*----------------------------------------------------------------------------
+| Functions and definitions to determine:  (1) whether tininess for underflow
+| is detected before or after rounding by default, (2) what (if anything)
+| happens when exceptions are raised, (3) how signaling NaNs are distinguished
+| from quiet NaNs, (4) the default generated quiet NaNs, and (5) how NaNs
+| are propagated from function inputs to output.  These details are target-
+| specific.
+*----------------------------------------------------------------------------*/
+#include "softfloat-specialize.h"
+
 /* Canonicalize EXP and FRAC, setting CLS.  */
 static FloatParts canonicalize(FloatParts part, const FloatFmt *parm,
                                float_status *status)
-- 
2.17.0

[Qemu-devel] [PATCH v5 05/28] fpu/softfloat: Canonicalize NaN fraction

[Richard Henderson]

Shift the NaN fraction to a canonical position, much like we
do for the fraction of normal numbers.  This will facilitate
manipulation of NaNs within the shared code paths.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 0d17027379..607c4a78d5 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -330,10 +330,11 @@ static FloatParts canonicalize(FloatParts part, const FloatFmt *parm,
         if (part.frac == 0) {
             part.cls = float_class_inf;
         } else {
+            part.frac <<= parm->frac_shift;
 #ifdef NO_SIGNALING_NANS
             part.cls = float_class_qnan;
 #else
-            int64_t msb = part.frac << (parm->frac_shift + 2);
+            int64_t msb = part.frac << 2;
             if ((msb < 0) == status->snan_bit_is_one) {
                 part.cls = float_class_snan;
             } else {
@@ -480,6 +481,7 @@ static FloatParts round_canonical(FloatParts p, float_status *s,
     case float_class_qnan:
     case float_class_snan:
         exp = exp_max;
+        frac >>= parm->frac_shift;
         break;
 
     default:
@@ -503,6 +505,7 @@ static float16 float16_round_pack_canonical(FloatParts p, float_status *s)
     case float_class_dnan:
         return float16_default_nan(s);
     case float_class_msnan:
+        p.frac >>= float16_params.frac_shift;
         return float16_maybe_silence_nan(float16_pack_raw(p), s);
     default:
         p = round_canonical(p, s, &float16_params);
@@ -521,6 +524,7 @@ static float32 float32_round_pack_canonical(FloatParts p, float_status *s)
     case float_class_dnan:
         return float32_default_nan(s);
     case float_class_msnan:
+        p.frac >>= float32_params.frac_shift;
         return float32_maybe_silence_nan(float32_pack_raw(p), s);
     default:
         p = round_canonical(p, s, &float32_params);
@@ -539,6 +543,7 @@ static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
     case float_class_dnan:
         return float64_default_nan(s);
     case float_class_msnan:
+        p.frac >>= float64_params.frac_shift;
         return float64_maybe_silence_nan(float64_pack_raw(p), s);
     default:
         p = round_canonical(p, s, &float64_params);
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 05/28] fpu/softfloat: Canonicalize NaN fraction

[Peter Maydell]

On 14 May 2018 at 23:11, Richard Henderson <richard.henderson@linaro.org> wrote:
> Shift the NaN fraction to a canonical position, much like we
> do for the fraction of normal numbers.  This will facilitate
> manipulation of NaNs within the shared code paths.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

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

thanks
-- PMM

Re: [Qemu-devel] [PATCH v5 05/28] fpu/softfloat: Canonicalize NaN fraction

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> Shift the NaN fraction to a canonical position, much like we
> do for the fraction of normal numbers.  This will facilitate
> manipulation of NaNs within the shared code paths.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index 0d17027379..607c4a78d5 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -330,10 +330,11 @@ static FloatParts canonicalize(FloatParts part, const FloatFmt *parm,
>          if (part.frac == 0) {
>              part.cls = float_class_inf;
>          } else {
> +            part.frac <<= parm->frac_shift;
>  #ifdef NO_SIGNALING_NANS
>              part.cls = float_class_qnan;
>  #else
> -            int64_t msb = part.frac << (parm->frac_shift + 2);
> +            int64_t msb = part.frac << 2;
>              if ((msb < 0) == status->snan_bit_is_one) {
>                  part.cls = float_class_snan;
>              } else {
> @@ -480,6 +481,7 @@ static FloatParts round_canonical(FloatParts p, float_status *s,
>      case float_class_qnan:
>      case float_class_snan:
>          exp = exp_max;
> +        frac >>= parm->frac_shift;
>          break;
>
>      default:
> @@ -503,6 +505,7 @@ static float16 float16_round_pack_canonical(FloatParts p, float_status *s)
>      case float_class_dnan:
>          return float16_default_nan(s);
>      case float_class_msnan:
> +        p.frac >>= float16_params.frac_shift;
>          return float16_maybe_silence_nan(float16_pack_raw(p), s);
>      default:
>          p = round_canonical(p, s, &float16_params);
> @@ -521,6 +524,7 @@ static float32 float32_round_pack_canonical(FloatParts p, float_status *s)
>      case float_class_dnan:
>          return float32_default_nan(s);
>      case float_class_msnan:
> +        p.frac >>= float32_params.frac_shift;
>          return float32_maybe_silence_nan(float32_pack_raw(p), s);
>      default:
>          p = round_canonical(p, s, &float32_params);
> @@ -539,6 +543,7 @@ static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
>      case float_class_dnan:
>          return float64_default_nan(s);
>      case float_class_msnan:
> +        p.frac >>= float64_params.frac_shift;
>          return float64_maybe_silence_nan(float64_pack_raw(p), s);
>      default:
>          p = round_canonical(p, s, &float64_params);


--
Alex Bennée

[Qemu-devel] [PATCH v5 06/28] fpu/softfloat: Introduce parts_is_snan_frac

[Richard Henderson]

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 15 +++++++++++++++
 fpu/softfloat.c            | 12 ++----------
 2 files changed, 17 insertions(+), 10 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 4fc9ea4ac0..515cb12cfa 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -86,6 +86,21 @@ this code that are retained.
 #define NO_SIGNALING_NANS 1
 #endif
 
+/*----------------------------------------------------------------------------
+| For the deconstructed floating-point with fraction FRAC, return true
+| if the fraction represents a signalling NaN; otherwise false.
+*----------------------------------------------------------------------------*/
+
+static bool parts_is_snan_frac(uint64_t frac, float_status *status)
+{
+#ifdef NO_SIGNALING_NANS
+    return false;
+#else
+    flag msb = extract64(frac, DECOMPOSED_BINARY_POINT - 1, 1);
+    return msb == status->snan_bit_is_one;
+#endif
+}
+
 /*----------------------------------------------------------------------------
 | The pattern for a default generated half-precision NaN.
 *----------------------------------------------------------------------------*/
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 607c4a78d5..19f40d6932 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -331,16 +331,8 @@ static FloatParts canonicalize(FloatParts part, const FloatFmt *parm,
             part.cls = float_class_inf;
         } else {
             part.frac <<= parm->frac_shift;
-#ifdef NO_SIGNALING_NANS
-            part.cls = float_class_qnan;
-#else
-            int64_t msb = part.frac << 2;
-            if ((msb < 0) == status->snan_bit_is_one) {
-                part.cls = float_class_snan;
-            } else {
-                part.cls = float_class_qnan;
-            }
-#endif
+            part.cls = (parts_is_snan_frac(part.frac, status)
+                        ? float_class_snan : float_class_qnan);
         }
     } else if (part.exp == 0) {
         if (likely(part.frac == 0)) {
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 06/28] fpu/softfloat: Introduce parts_is_snan_frac

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

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

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat-specialize.h | 15 +++++++++++++++
>  fpu/softfloat.c            | 12 ++----------
>  2 files changed, 17 insertions(+), 10 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index 4fc9ea4ac0..515cb12cfa 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -86,6 +86,21 @@ this code that are retained.
>  #define NO_SIGNALING_NANS 1
>  #endif
>
> +/*----------------------------------------------------------------------------
> +| For the deconstructed floating-point with fraction FRAC, return true
> +| if the fraction represents a signalling NaN; otherwise false.
> +*----------------------------------------------------------------------------*/
> +
> +static bool parts_is_snan_frac(uint64_t frac, float_status *status)
> +{
> +#ifdef NO_SIGNALING_NANS
> +    return false;
> +#else
> +    flag msb = extract64(frac, DECOMPOSED_BINARY_POINT - 1, 1);
> +    return msb == status->snan_bit_is_one;
> +#endif
> +}
> +
>  /*----------------------------------------------------------------------------
>  | The pattern for a default generated half-precision NaN.
>  *----------------------------------------------------------------------------*/
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index 607c4a78d5..19f40d6932 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -331,16 +331,8 @@ static FloatParts canonicalize(FloatParts part, const FloatFmt *parm,
>              part.cls = float_class_inf;
>          } else {
>              part.frac <<= parm->frac_shift;
> -#ifdef NO_SIGNALING_NANS
> -            part.cls = float_class_qnan;
> -#else
> -            int64_t msb = part.frac << 2;
> -            if ((msb < 0) == status->snan_bit_is_one) {
> -                part.cls = float_class_snan;
> -            } else {
> -                part.cls = float_class_qnan;
> -            }
> -#endif
> +            part.cls = (parts_is_snan_frac(part.frac, status)
> +                        ? float_class_snan : float_class_qnan);
>          }
>      } else if (part.exp == 0) {
>          if (likely(part.frac == 0)) {


--
Alex Bennée

[Qemu-devel] [PATCH v5 07/28] fpu/softfloat: Replace float_class_dnan with parts_default_nan

[Richard Henderson]

With a canonical representation of NaNs, we can return the
default nan directly rather than delay the expansion until
the final format is known.

Note one case where we uselessly assigned to a.sign, which was
overwritten/ignored later when expanding float_class_dnan.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 37 +++++++++++++++++++++++++++++++++++++
 fpu/softfloat.c            | 38 +++++++++++---------------------------
 2 files changed, 48 insertions(+), 27 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 515cb12cfa..0d3d81a52b 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -101,6 +101,43 @@ static bool parts_is_snan_frac(uint64_t frac, float_status *status)
 #endif
 }
 
+/*----------------------------------------------------------------------------
+| The pattern for a default generated deconstructed floating-point NaN.
+*----------------------------------------------------------------------------*/
+
+static FloatParts parts_default_nan(float_status *status)
+{
+    bool sign = 0;
+    uint64_t frac;
+
+#if defined(TARGET_SPARC) || defined(TARGET_M68K)
+    frac = (1ULL << DECOMPOSED_BINARY_POINT) - 1;
+#elif defined(TARGET_PPC) || defined(TARGET_ARM) || defined(TARGET_ALPHA) || \
+      defined(TARGET_S390X) || defined(TARGET_RISCV)
+    frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
+#elif defined(TARGET_HPPA)
+    frac = 1ULL << (DECOMPOSED_BINARY_POINT - 2);
+#else
+    if (status->snan_bit_is_one) {
+        frac = (1ULL << (DECOMPOSED_BINARY_POINT - 1)) - 1;
+    } else {
+#if defined(TARGET_MIPS)
+        frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
+#else
+        frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
+        sign = 1;
+#endif
+    }
+#endif
+
+    return (FloatParts) {
+        .cls = float_class_qnan,
+        .sign = sign,
+        .exp = INT_MAX,
+        .frac = frac
+    };
+}
+
 /*----------------------------------------------------------------------------
 | The pattern for a default generated half-precision NaN.
 *----------------------------------------------------------------------------*/
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 19f40d6932..51780b718f 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -188,7 +188,6 @@ typedef enum __attribute__ ((__packed__)) {
     float_class_inf,
     float_class_qnan,  /* all NaNs from here */
     float_class_snan,
-    float_class_dnan,
     float_class_msnan, /* maybe silenced */
 } FloatClass;
 
@@ -494,8 +493,6 @@ static FloatParts float16_unpack_canonical(float16 f, float_status *s)
 static float16 float16_round_pack_canonical(FloatParts p, float_status *s)
 {
     switch (p.cls) {
-    case float_class_dnan:
-        return float16_default_nan(s);
     case float_class_msnan:
         p.frac >>= float16_params.frac_shift;
         return float16_maybe_silence_nan(float16_pack_raw(p), s);
@@ -513,8 +510,6 @@ static FloatParts float32_unpack_canonical(float32 f, float_status *s)
 static float32 float32_round_pack_canonical(FloatParts p, float_status *s)
 {
     switch (p.cls) {
-    case float_class_dnan:
-        return float32_default_nan(s);
     case float_class_msnan:
         p.frac >>= float32_params.frac_shift;
         return float32_maybe_silence_nan(float32_pack_raw(p), s);
@@ -532,8 +527,6 @@ static FloatParts float64_unpack_canonical(float64 f, float_status *s)
 static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
 {
     switch (p.cls) {
-    case float_class_dnan:
-        return float64_default_nan(s);
     case float_class_msnan:
         p.frac >>= float64_params.frac_shift;
         return float64_maybe_silence_nan(float64_pack_raw(p), s);
@@ -566,7 +559,7 @@ static FloatParts return_nan(FloatParts a, float_status *s)
         /* fall through */
     case float_class_qnan:
         if (s->default_nan_mode) {
-            a.cls = float_class_dnan;
+            return parts_default_nan(s);
         }
         break;
 
@@ -583,7 +576,7 @@ static FloatParts pick_nan(FloatParts a, FloatParts b, float_status *s)
     }
 
     if (s->default_nan_mode) {
-        a.cls = float_class_dnan;
+        return parts_default_nan(s);
     } else {
         if (pickNaN(is_qnan(a.cls), is_snan(a.cls),
                     is_qnan(b.cls), is_snan(b.cls),
@@ -614,8 +607,7 @@ static FloatParts pick_nan_muladd(FloatParts a, FloatParts b, FloatParts c,
         /* Note that this check is after pickNaNMulAdd so that function
          * has an opportunity to set the Invalid flag.
          */
-        a.cls = float_class_dnan;
-        return a;
+        which = 3;
     }
 
     switch (which) {
@@ -628,8 +620,7 @@ static FloatParts pick_nan_muladd(FloatParts a, FloatParts b, FloatParts c,
         a = c;
         break;
     case 3:
-        a.cls = float_class_dnan;
-        return a;
+        return parts_default_nan(s);
     default:
         g_assert_not_reached();
     }
@@ -682,7 +673,7 @@ static FloatParts addsub_floats(FloatParts a, FloatParts b, bool subtract,
         if (a.cls == float_class_inf) {
             if (b.cls == float_class_inf) {
                 float_raise(float_flag_invalid, s);
-                a.cls = float_class_dnan;
+                return parts_default_nan(s);
             }
             return a;
         }
@@ -828,9 +819,7 @@ static FloatParts mul_floats(FloatParts a, FloatParts b, float_status *s)
     if ((a.cls == float_class_inf && b.cls == float_class_zero) ||
         (a.cls == float_class_zero && b.cls == float_class_inf)) {
         s->float_exception_flags |= float_flag_invalid;
-        a.cls = float_class_dnan;
-        a.sign = sign;
-        return a;
+        return parts_default_nan(s);
     }
     /* Multiply by 0 or Inf */
     if (a.cls == float_class_inf || a.cls == float_class_zero) {
@@ -908,8 +897,7 @@ static FloatParts muladd_floats(FloatParts a, FloatParts b, FloatParts c,
 
     if (inf_zero) {
         s->float_exception_flags |= float_flag_invalid;
-        a.cls = float_class_dnan;
-        return a;
+        return parts_default_nan(s);
     }
 
     if (flags & float_muladd_negate_c) {
@@ -933,12 +921,12 @@ static FloatParts muladd_floats(FloatParts a, FloatParts b, FloatParts c,
     if (c.cls == float_class_inf) {
         if (p_class == float_class_inf && p_sign != c.sign) {
             s->float_exception_flags |= float_flag_invalid;
-            a.cls = float_class_dnan;
+            return parts_default_nan(s);
         } else {
             a.cls = float_class_inf;
             a.sign = c.sign ^ sign_flip;
+            return a;
         }
-        return a;
     }
 
     if (p_class == float_class_inf) {
@@ -1148,8 +1136,7 @@ static FloatParts div_floats(FloatParts a, FloatParts b, float_status *s)
         &&
         (a.cls == float_class_inf || a.cls == float_class_zero)) {
         s->float_exception_flags |= float_flag_invalid;
-        a.cls = float_class_dnan;
-        return a;
+        return parts_default_nan(s);
     }
     /* Inf / x or 0 / x */
     if (a.cls == float_class_inf || a.cls == float_class_zero) {
@@ -1347,7 +1334,6 @@ static int64_t round_to_int_and_pack(FloatParts in, int rmode,
     switch (p.cls) {
     case float_class_snan:
     case float_class_qnan:
-    case float_class_dnan:
     case float_class_msnan:
         s->float_exception_flags = orig_flags | float_flag_invalid;
         return max;
@@ -1439,7 +1425,6 @@ static uint64_t round_to_uint_and_pack(FloatParts in, int rmode, uint64_t max,
     switch (p.cls) {
     case float_class_snan:
     case float_class_qnan:
-    case float_class_dnan:
     case float_class_msnan:
         s->float_exception_flags = orig_flags | float_flag_invalid;
         return max;
@@ -1940,8 +1925,7 @@ static FloatParts sqrt_float(FloatParts a, float_status *s, const FloatFmt *p)
     }
     if (a.sign) {
         s->float_exception_flags |= float_flag_invalid;
-        a.cls = float_class_dnan;
-        return a;
+        return parts_default_nan(s);
     }
     if (a.cls == float_class_inf) {
         return a;  /* sqrt(+inf) = +inf */
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 07/28] fpu/softfloat: Replace float_class_dnan with parts_default_nan

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> With a canonical representation of NaNs, we can return the
> default nan directly rather than delay the expansion until
> the final format is known.
>
> Note one case where we uselessly assigned to a.sign, which was
> overwritten/ignored later when expanding float_class_dnan.
>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat-specialize.h | 37 +++++++++++++++++++++++++++++++++++++
>  fpu/softfloat.c            | 38 +++++++++++---------------------------
>  2 files changed, 48 insertions(+), 27 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index 515cb12cfa..0d3d81a52b 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -101,6 +101,43 @@ static bool parts_is_snan_frac(uint64_t frac, float_status *status)
>  #endif
>  }
>
> +/*----------------------------------------------------------------------------
> +| The pattern for a default generated deconstructed floating-point NaN.
> +*----------------------------------------------------------------------------*/
> +
> +static FloatParts parts_default_nan(float_status *status)
> +{
> +    bool sign = 0;
> +    uint64_t frac;
> +
> +#if defined(TARGET_SPARC) || defined(TARGET_M68K)
> +    frac = (1ULL << DECOMPOSED_BINARY_POINT) - 1;
> +#elif defined(TARGET_PPC) || defined(TARGET_ARM) || defined(TARGET_ALPHA) || \
> +      defined(TARGET_S390X) || defined(TARGET_RISCV)
> +    frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> +#elif defined(TARGET_HPPA)
> +    frac = 1ULL << (DECOMPOSED_BINARY_POINT - 2);
> +#else
> +    if (status->snan_bit_is_one) {
> +        frac = (1ULL << (DECOMPOSED_BINARY_POINT - 1)) - 1;
> +    } else {
> +#if defined(TARGET_MIPS)
> +        frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> +#else
> +        frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> +        sign = 1;
> +#endif
> +    }
> +#endif
> +
> +    return (FloatParts) {
> +        .cls = float_class_qnan,
> +        .sign = sign,
> +        .exp = INT_MAX,
> +        .frac = frac
> +    };
> +}
> +
>  /*----------------------------------------------------------------------------
>  | The pattern for a default generated half-precision NaN.
>  *----------------------------------------------------------------------------*/
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index 19f40d6932..51780b718f 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -188,7 +188,6 @@ typedef enum __attribute__ ((__packed__)) {
>      float_class_inf,
>      float_class_qnan,  /* all NaNs from here */
>      float_class_snan,
> -    float_class_dnan,
>      float_class_msnan, /* maybe silenced */
>  } FloatClass;
>
> @@ -494,8 +493,6 @@ static FloatParts float16_unpack_canonical(float16 f, float_status *s)
>  static float16 float16_round_pack_canonical(FloatParts p, float_status *s)
>  {
>      switch (p.cls) {
> -    case float_class_dnan:
> -        return float16_default_nan(s);
>      case float_class_msnan:
>          p.frac >>= float16_params.frac_shift;
>          return float16_maybe_silence_nan(float16_pack_raw(p), s);
> @@ -513,8 +510,6 @@ static FloatParts float32_unpack_canonical(float32 f, float_status *s)
>  static float32 float32_round_pack_canonical(FloatParts p, float_status *s)
>  {
>      switch (p.cls) {
> -    case float_class_dnan:
> -        return float32_default_nan(s);
>      case float_class_msnan:
>          p.frac >>= float32_params.frac_shift;
>          return float32_maybe_silence_nan(float32_pack_raw(p), s);
> @@ -532,8 +527,6 @@ static FloatParts float64_unpack_canonical(float64 f, float_status *s)
>  static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
>  {
>      switch (p.cls) {
> -    case float_class_dnan:
> -        return float64_default_nan(s);
>      case float_class_msnan:
>          p.frac >>= float64_params.frac_shift;
>          return float64_maybe_silence_nan(float64_pack_raw(p), s);
> @@ -566,7 +559,7 @@ static FloatParts return_nan(FloatParts a, float_status *s)
>          /* fall through */
>      case float_class_qnan:
>          if (s->default_nan_mode) {
> -            a.cls = float_class_dnan;
> +            return parts_default_nan(s);
>          }
>          break;
>
> @@ -583,7 +576,7 @@ static FloatParts pick_nan(FloatParts a, FloatParts b, float_status *s)
>      }
>
>      if (s->default_nan_mode) {
> -        a.cls = float_class_dnan;
> +        return parts_default_nan(s);
>      } else {
>          if (pickNaN(is_qnan(a.cls), is_snan(a.cls),
>                      is_qnan(b.cls), is_snan(b.cls),
> @@ -614,8 +607,7 @@ static FloatParts pick_nan_muladd(FloatParts a, FloatParts b, FloatParts c,
>          /* Note that this check is after pickNaNMulAdd so that function
>           * has an opportunity to set the Invalid flag.
>           */
> -        a.cls = float_class_dnan;
> -        return a;
> +        which = 3;
>      }
>
>      switch (which) {
> @@ -628,8 +620,7 @@ static FloatParts pick_nan_muladd(FloatParts a, FloatParts b, FloatParts c,
>          a = c;
>          break;
>      case 3:
> -        a.cls = float_class_dnan;
> -        return a;
> +        return parts_default_nan(s);
>      default:
>          g_assert_not_reached();
>      }
> @@ -682,7 +673,7 @@ static FloatParts addsub_floats(FloatParts a, FloatParts b, bool subtract,
>          if (a.cls == float_class_inf) {
>              if (b.cls == float_class_inf) {
>                  float_raise(float_flag_invalid, s);
> -                a.cls = float_class_dnan;
> +                return parts_default_nan(s);
>              }
>              return a;
>          }
> @@ -828,9 +819,7 @@ static FloatParts mul_floats(FloatParts a, FloatParts b, float_status *s)
>      if ((a.cls == float_class_inf && b.cls == float_class_zero) ||
>          (a.cls == float_class_zero && b.cls == float_class_inf)) {
>          s->float_exception_flags |= float_flag_invalid;
> -        a.cls = float_class_dnan;
> -        a.sign = sign;
> -        return a;
> +        return parts_default_nan(s);
>      }
>      /* Multiply by 0 or Inf */
>      if (a.cls == float_class_inf || a.cls == float_class_zero) {
> @@ -908,8 +897,7 @@ static FloatParts muladd_floats(FloatParts a, FloatParts b, FloatParts c,
>
>      if (inf_zero) {
>          s->float_exception_flags |= float_flag_invalid;
> -        a.cls = float_class_dnan;
> -        return a;
> +        return parts_default_nan(s);
>      }
>
>      if (flags & float_muladd_negate_c) {
> @@ -933,12 +921,12 @@ static FloatParts muladd_floats(FloatParts a, FloatParts b, FloatParts c,
>      if (c.cls == float_class_inf) {
>          if (p_class == float_class_inf && p_sign != c.sign) {
>              s->float_exception_flags |= float_flag_invalid;
> -            a.cls = float_class_dnan;
> +            return parts_default_nan(s);
>          } else {
>              a.cls = float_class_inf;
>              a.sign = c.sign ^ sign_flip;
> +            return a;
>          }
> -        return a;
>      }
>
>      if (p_class == float_class_inf) {
> @@ -1148,8 +1136,7 @@ static FloatParts div_floats(FloatParts a, FloatParts b, float_status *s)
>          &&
>          (a.cls == float_class_inf || a.cls == float_class_zero)) {
>          s->float_exception_flags |= float_flag_invalid;
> -        a.cls = float_class_dnan;
> -        return a;
> +        return parts_default_nan(s);
>      }
>      /* Inf / x or 0 / x */
>      if (a.cls == float_class_inf || a.cls == float_class_zero) {
> @@ -1347,7 +1334,6 @@ static int64_t round_to_int_and_pack(FloatParts in, int rmode,
>      switch (p.cls) {
>      case float_class_snan:
>      case float_class_qnan:
> -    case float_class_dnan:
>      case float_class_msnan:
>          s->float_exception_flags = orig_flags | float_flag_invalid;
>          return max;
> @@ -1439,7 +1425,6 @@ static uint64_t round_to_uint_and_pack(FloatParts in, int rmode, uint64_t max,
>      switch (p.cls) {
>      case float_class_snan:
>      case float_class_qnan:
> -    case float_class_dnan:
>      case float_class_msnan:
>          s->float_exception_flags = orig_flags | float_flag_invalid;
>          return max;
> @@ -1940,8 +1925,7 @@ static FloatParts sqrt_float(FloatParts a, float_status *s, const FloatFmt *p)
>      }
>      if (a.sign) {
>          s->float_exception_flags |= float_flag_invalid;
> -        a.cls = float_class_dnan;
> -        return a;
> +        return parts_default_nan(s);
>      }
>      if (a.cls == float_class_inf) {
>          return a;  /* sqrt(+inf) = +inf */


--
Alex Bennée

[Qemu-devel] [PATCH v5 08/28] fpu/softfloat: Replace float_class_msnan with parts_silence_nan

[Richard Henderson]

With a canonical representation of NaNs, we can silence an SNaN
immediately rather than delay until the final format is known.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 23 ++++++++++++++++++++++
 fpu/softfloat.c            | 40 ++++++++++----------------------------
 2 files changed, 33 insertions(+), 30 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 0d3d81a52b..571d1df378 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -138,6 +138,29 @@ static FloatParts parts_default_nan(float_status *status)
     };
 }
 
+/*----------------------------------------------------------------------------
+| Returns a quiet NaN from a signalling NaN for the deconstructed
+| floating-point parts.
+*----------------------------------------------------------------------------*/
+
+static FloatParts parts_silence_nan(FloatParts a, float_status *status)
+{
+#ifdef NO_SIGNALING_NANS
+    g_assert_not_reached();
+#elif defined(TARGET_HPPA)
+    a.frac &= ~(1ULL << (DECOMPOSED_BINARY_POINT - 1));
+    a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 2);
+#else
+    if (status->snan_bit_is_one) {
+        return parts_default_nan(status);
+    } else {
+        a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 1);
+    }
+#endif
+    a.cls = float_class_qnan;
+    return a;
+}
+
 /*----------------------------------------------------------------------------
 | The pattern for a default generated half-precision NaN.
 *----------------------------------------------------------------------------*/
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 51780b718f..41253c6749 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -188,7 +188,6 @@ typedef enum __attribute__ ((__packed__)) {
     float_class_inf,
     float_class_qnan,  /* all NaNs from here */
     float_class_snan,
-    float_class_msnan, /* maybe silenced */
 } FloatClass;
 
 /*
@@ -492,14 +491,7 @@ static FloatParts float16_unpack_canonical(float16 f, float_status *s)
 
 static float16 float16_round_pack_canonical(FloatParts p, float_status *s)
 {
-    switch (p.cls) {
-    case float_class_msnan:
-        p.frac >>= float16_params.frac_shift;
-        return float16_maybe_silence_nan(float16_pack_raw(p), s);
-    default:
-        p = round_canonical(p, s, &float16_params);
-        return float16_pack_raw(p);
-    }
+    return float16_pack_raw(round_canonical(p, s, &float16_params));
 }
 
 static FloatParts float32_unpack_canonical(float32 f, float_status *s)
@@ -509,14 +501,7 @@ static FloatParts float32_unpack_canonical(float32 f, float_status *s)
 
 static float32 float32_round_pack_canonical(FloatParts p, float_status *s)
 {
-    switch (p.cls) {
-    case float_class_msnan:
-        p.frac >>= float32_params.frac_shift;
-        return float32_maybe_silence_nan(float32_pack_raw(p), s);
-    default:
-        p = round_canonical(p, s, &float32_params);
-        return float32_pack_raw(p);
-    }
+    return float32_pack_raw(round_canonical(p, s, &float32_params));
 }
 
 static FloatParts float64_unpack_canonical(float64 f, float_status *s)
@@ -526,14 +511,7 @@ static FloatParts float64_unpack_canonical(float64 f, float_status *s)
 
 static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
 {
-    switch (p.cls) {
-    case float_class_msnan:
-        p.frac >>= float64_params.frac_shift;
-        return float64_maybe_silence_nan(float64_pack_raw(p), s);
-    default:
-        p = round_canonical(p, s, &float64_params);
-        return float64_pack_raw(p);
-    }
+    return float64_pack_raw(round_canonical(p, s, &float64_params));
 }
 
 /* Simple helpers for checking if what NaN we have */
@@ -555,7 +533,7 @@ static FloatParts return_nan(FloatParts a, float_status *s)
     switch (a.cls) {
     case float_class_snan:
         s->float_exception_flags |= float_flag_invalid;
-        a.cls = float_class_msnan;
+        a = parts_silence_nan(a, s);
         /* fall through */
     case float_class_qnan:
         if (s->default_nan_mode) {
@@ -584,7 +562,9 @@ static FloatParts pick_nan(FloatParts a, FloatParts b, float_status *s)
                     (a.frac == b.frac && a.sign < b.sign))) {
             a = b;
         }
-        a.cls = float_class_msnan;
+        if (is_snan(a.cls)) {
+            return parts_silence_nan(a, s);
+        }
     }
     return a;
 }
@@ -624,8 +604,10 @@ static FloatParts pick_nan_muladd(FloatParts a, FloatParts b, FloatParts c,
     default:
         g_assert_not_reached();
     }
-    a.cls = float_class_msnan;
 
+    if (is_snan(a.cls)) {
+        return parts_silence_nan(a, s);
+    }
     return a;
 }
 
@@ -1334,7 +1316,6 @@ static int64_t round_to_int_and_pack(FloatParts in, int rmode,
     switch (p.cls) {
     case float_class_snan:
     case float_class_qnan:
-    case float_class_msnan:
         s->float_exception_flags = orig_flags | float_flag_invalid;
         return max;
     case float_class_inf:
@@ -1425,7 +1406,6 @@ static uint64_t round_to_uint_and_pack(FloatParts in, int rmode, uint64_t max,
     switch (p.cls) {
     case float_class_snan:
     case float_class_qnan:
-    case float_class_msnan:
         s->float_exception_flags = orig_flags | float_flag_invalid;
         return max;
     case float_class_inf:
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 08/28] fpu/softfloat: Replace float_class_msnan with parts_silence_nan

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> With a canonical representation of NaNs, we can silence an SNaN
> immediately rather than delay until the final format is known.
>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat-specialize.h | 23 ++++++++++++++++++++++
>  fpu/softfloat.c            | 40 ++++++++++----------------------------
>  2 files changed, 33 insertions(+), 30 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index 0d3d81a52b..571d1df378 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -138,6 +138,29 @@ static FloatParts parts_default_nan(float_status *status)
>      };
>  }
>
> +/*----------------------------------------------------------------------------
> +| Returns a quiet NaN from a signalling NaN for the deconstructed
> +| floating-point parts.
> +*----------------------------------------------------------------------------*/
> +
> +static FloatParts parts_silence_nan(FloatParts a, float_status *status)
> +{
> +#ifdef NO_SIGNALING_NANS
> +    g_assert_not_reached();
> +#elif defined(TARGET_HPPA)
> +    a.frac &= ~(1ULL << (DECOMPOSED_BINARY_POINT - 1));
> +    a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 2);
> +#else
> +    if (status->snan_bit_is_one) {
> +        return parts_default_nan(status);
> +    } else {
> +        a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> +    }
> +#endif
> +    a.cls = float_class_qnan;
> +    return a;
> +}
> +
>  /*----------------------------------------------------------------------------
>  | The pattern for a default generated half-precision NaN.
>  *----------------------------------------------------------------------------*/
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index 51780b718f..41253c6749 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -188,7 +188,6 @@ typedef enum __attribute__ ((__packed__)) {
>      float_class_inf,
>      float_class_qnan,  /* all NaNs from here */
>      float_class_snan,
> -    float_class_msnan, /* maybe silenced */
>  } FloatClass;
>
>  /*
> @@ -492,14 +491,7 @@ static FloatParts float16_unpack_canonical(float16 f, float_status *s)
>
>  static float16 float16_round_pack_canonical(FloatParts p, float_status *s)
>  {
> -    switch (p.cls) {
> -    case float_class_msnan:
> -        p.frac >>= float16_params.frac_shift;
> -        return float16_maybe_silence_nan(float16_pack_raw(p), s);
> -    default:
> -        p = round_canonical(p, s, &float16_params);
> -        return float16_pack_raw(p);
> -    }
> +    return float16_pack_raw(round_canonical(p, s, &float16_params));
>  }
>
>  static FloatParts float32_unpack_canonical(float32 f, float_status *s)
> @@ -509,14 +501,7 @@ static FloatParts float32_unpack_canonical(float32 f, float_status *s)
>
>  static float32 float32_round_pack_canonical(FloatParts p, float_status *s)
>  {
> -    switch (p.cls) {
> -    case float_class_msnan:
> -        p.frac >>= float32_params.frac_shift;
> -        return float32_maybe_silence_nan(float32_pack_raw(p), s);
> -    default:
> -        p = round_canonical(p, s, &float32_params);
> -        return float32_pack_raw(p);
> -    }
> +    return float32_pack_raw(round_canonical(p, s, &float32_params));
>  }
>
>  static FloatParts float64_unpack_canonical(float64 f, float_status *s)
> @@ -526,14 +511,7 @@ static FloatParts float64_unpack_canonical(float64 f, float_status *s)
>
>  static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
>  {
> -    switch (p.cls) {
> -    case float_class_msnan:
> -        p.frac >>= float64_params.frac_shift;
> -        return float64_maybe_silence_nan(float64_pack_raw(p), s);
> -    default:
> -        p = round_canonical(p, s, &float64_params);
> -        return float64_pack_raw(p);
> -    }
> +    return float64_pack_raw(round_canonical(p, s, &float64_params));
>  }
>
>  /* Simple helpers for checking if what NaN we have */
> @@ -555,7 +533,7 @@ static FloatParts return_nan(FloatParts a, float_status *s)
>      switch (a.cls) {
>      case float_class_snan:
>          s->float_exception_flags |= float_flag_invalid;
> -        a.cls = float_class_msnan;
> +        a = parts_silence_nan(a, s);
>          /* fall through */
>      case float_class_qnan:
>          if (s->default_nan_mode) {
> @@ -584,7 +562,9 @@ static FloatParts pick_nan(FloatParts a, FloatParts b, float_status *s)
>                      (a.frac == b.frac && a.sign < b.sign))) {
>              a = b;
>          }
> -        a.cls = float_class_msnan;
> +        if (is_snan(a.cls)) {
> +            return parts_silence_nan(a, s);
> +        }
>      }
>      return a;
>  }
> @@ -624,8 +604,10 @@ static FloatParts pick_nan_muladd(FloatParts a, FloatParts b, FloatParts c,
>      default:
>          g_assert_not_reached();
>      }
> -    a.cls = float_class_msnan;
>
> +    if (is_snan(a.cls)) {
> +        return parts_silence_nan(a, s);
> +    }
>      return a;
>  }
>
> @@ -1334,7 +1316,6 @@ static int64_t round_to_int_and_pack(FloatParts in, int rmode,
>      switch (p.cls) {
>      case float_class_snan:
>      case float_class_qnan:
> -    case float_class_msnan:
>          s->float_exception_flags = orig_flags | float_flag_invalid;
>          return max;
>      case float_class_inf:
> @@ -1425,7 +1406,6 @@ static uint64_t round_to_uint_and_pack(FloatParts in, int rmode, uint64_t max,
>      switch (p.cls) {
>      case float_class_snan:
>      case float_class_qnan:
> -    case float_class_msnan:
>          s->float_exception_flags = orig_flags | float_flag_invalid;
>          return max;
>      case float_class_inf:


--
Alex Bennée

[Qemu-devel] [PATCH v5 09/28] target/arm: Use floatX_silence_nan when we have already checked for SNaN

[Richard Henderson]

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/helper-a64.c |  6 +++---
 target/arm/helper.c     | 12 ++++++------
 2 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/target/arm/helper-a64.c b/target/arm/helper-a64.c
index 4f8034c513..6f0eb83661 100644
--- a/target/arm/helper-a64.c
+++ b/target/arm/helper-a64.c
@@ -376,7 +376,7 @@ float16 HELPER(frecpx_f16)(float16 a, void *fpstp)
         float16 nan = a;
         if (float16_is_signaling_nan(a, fpst)) {
             float_raise(float_flag_invalid, fpst);
-            nan = float16_maybe_silence_nan(a, fpst);
+            nan = float16_silence_nan(a, fpst);
         }
         if (fpst->default_nan_mode) {
             nan = float16_default_nan(fpst);
@@ -405,7 +405,7 @@ float32 HELPER(frecpx_f32)(float32 a, void *fpstp)
         float32 nan = a;
         if (float32_is_signaling_nan(a, fpst)) {
             float_raise(float_flag_invalid, fpst);
-            nan = float32_maybe_silence_nan(a, fpst);
+            nan = float32_silence_nan(a, fpst);
         }
         if (fpst->default_nan_mode) {
             nan = float32_default_nan(fpst);
@@ -434,7 +434,7 @@ float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
         float64 nan = a;
         if (float64_is_signaling_nan(a, fpst)) {
             float_raise(float_flag_invalid, fpst);
-            nan = float64_maybe_silence_nan(a, fpst);
+            nan = float64_silence_nan(a, fpst);
         }
         if (fpst->default_nan_mode) {
             nan = float64_default_nan(fpst);
diff --git a/target/arm/helper.c b/target/arm/helper.c
index c6fd7f9479..3df5cf30e4 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -11739,7 +11739,7 @@ float16 HELPER(recpe_f16)(float16 input, void *fpstp)
         float16 nan = f16;
         if (float16_is_signaling_nan(f16, fpst)) {
             float_raise(float_flag_invalid, fpst);
-            nan = float16_maybe_silence_nan(f16, fpst);
+            nan = float16_silence_nan(f16, fpst);
         }
         if (fpst->default_nan_mode) {
             nan =  float16_default_nan(fpst);
@@ -11787,7 +11787,7 @@ float32 HELPER(recpe_f32)(float32 input, void *fpstp)
         float32 nan = f32;
         if (float32_is_signaling_nan(f32, fpst)) {
             float_raise(float_flag_invalid, fpst);
-            nan = float32_maybe_silence_nan(f32, fpst);
+            nan = float32_silence_nan(f32, fpst);
         }
         if (fpst->default_nan_mode) {
             nan =  float32_default_nan(fpst);
@@ -11835,7 +11835,7 @@ float64 HELPER(recpe_f64)(float64 input, void *fpstp)
         float64 nan = f64;
         if (float64_is_signaling_nan(f64, fpst)) {
             float_raise(float_flag_invalid, fpst);
-            nan = float64_maybe_silence_nan(f64, fpst);
+            nan = float64_silence_nan(f64, fpst);
         }
         if (fpst->default_nan_mode) {
             nan =  float64_default_nan(fpst);
@@ -11934,7 +11934,7 @@ float16 HELPER(rsqrte_f16)(float16 input, void *fpstp)
         float16 nan = f16;
         if (float16_is_signaling_nan(f16, s)) {
             float_raise(float_flag_invalid, s);
-            nan = float16_maybe_silence_nan(f16, s);
+            nan = float16_silence_nan(f16, s);
         }
         if (s->default_nan_mode) {
             nan =  float16_default_nan(s);
@@ -11978,7 +11978,7 @@ float32 HELPER(rsqrte_f32)(float32 input, void *fpstp)
         float32 nan = f32;
         if (float32_is_signaling_nan(f32, s)) {
             float_raise(float_flag_invalid, s);
-            nan = float32_maybe_silence_nan(f32, s);
+            nan = float32_silence_nan(f32, s);
         }
         if (s->default_nan_mode) {
             nan =  float32_default_nan(s);
@@ -12021,7 +12021,7 @@ float64 HELPER(rsqrte_f64)(float64 input, void *fpstp)
         float64 nan = f64;
         if (float64_is_signaling_nan(f64, s)) {
             float_raise(float_flag_invalid, s);
-            nan = float64_maybe_silence_nan(f64, s);
+            nan = float64_silence_nan(f64, s);
         }
         if (s->default_nan_mode) {
             nan =  float64_default_nan(s);
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 09/28] target/arm: Use floatX_silence_nan when we have already checked for SNaN

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

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

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  target/arm/helper-a64.c |  6 +++---
>  target/arm/helper.c     | 12 ++++++------
>  2 files changed, 9 insertions(+), 9 deletions(-)
>
> diff --git a/target/arm/helper-a64.c b/target/arm/helper-a64.c
> index 4f8034c513..6f0eb83661 100644
> --- a/target/arm/helper-a64.c
> +++ b/target/arm/helper-a64.c
> @@ -376,7 +376,7 @@ float16 HELPER(frecpx_f16)(float16 a, void *fpstp)
>          float16 nan = a;
>          if (float16_is_signaling_nan(a, fpst)) {
>              float_raise(float_flag_invalid, fpst);
> -            nan = float16_maybe_silence_nan(a, fpst);
> +            nan = float16_silence_nan(a, fpst);
>          }
>          if (fpst->default_nan_mode) {
>              nan = float16_default_nan(fpst);
> @@ -405,7 +405,7 @@ float32 HELPER(frecpx_f32)(float32 a, void *fpstp)
>          float32 nan = a;
>          if (float32_is_signaling_nan(a, fpst)) {
>              float_raise(float_flag_invalid, fpst);
> -            nan = float32_maybe_silence_nan(a, fpst);
> +            nan = float32_silence_nan(a, fpst);
>          }
>          if (fpst->default_nan_mode) {
>              nan = float32_default_nan(fpst);
> @@ -434,7 +434,7 @@ float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
>          float64 nan = a;
>          if (float64_is_signaling_nan(a, fpst)) {
>              float_raise(float_flag_invalid, fpst);
> -            nan = float64_maybe_silence_nan(a, fpst);
> +            nan = float64_silence_nan(a, fpst);
>          }
>          if (fpst->default_nan_mode) {
>              nan = float64_default_nan(fpst);
> diff --git a/target/arm/helper.c b/target/arm/helper.c
> index c6fd7f9479..3df5cf30e4 100644
> --- a/target/arm/helper.c
> +++ b/target/arm/helper.c
> @@ -11739,7 +11739,7 @@ float16 HELPER(recpe_f16)(float16 input, void *fpstp)
>          float16 nan = f16;
>          if (float16_is_signaling_nan(f16, fpst)) {
>              float_raise(float_flag_invalid, fpst);
> -            nan = float16_maybe_silence_nan(f16, fpst);
> +            nan = float16_silence_nan(f16, fpst);
>          }
>          if (fpst->default_nan_mode) {
>              nan =  float16_default_nan(fpst);
> @@ -11787,7 +11787,7 @@ float32 HELPER(recpe_f32)(float32 input, void *fpstp)
>          float32 nan = f32;
>          if (float32_is_signaling_nan(f32, fpst)) {
>              float_raise(float_flag_invalid, fpst);
> -            nan = float32_maybe_silence_nan(f32, fpst);
> +            nan = float32_silence_nan(f32, fpst);
>          }
>          if (fpst->default_nan_mode) {
>              nan =  float32_default_nan(fpst);
> @@ -11835,7 +11835,7 @@ float64 HELPER(recpe_f64)(float64 input, void *fpstp)
>          float64 nan = f64;
>          if (float64_is_signaling_nan(f64, fpst)) {
>              float_raise(float_flag_invalid, fpst);
> -            nan = float64_maybe_silence_nan(f64, fpst);
> +            nan = float64_silence_nan(f64, fpst);
>          }
>          if (fpst->default_nan_mode) {
>              nan =  float64_default_nan(fpst);
> @@ -11934,7 +11934,7 @@ float16 HELPER(rsqrte_f16)(float16 input, void *fpstp)
>          float16 nan = f16;
>          if (float16_is_signaling_nan(f16, s)) {
>              float_raise(float_flag_invalid, s);
> -            nan = float16_maybe_silence_nan(f16, s);
> +            nan = float16_silence_nan(f16, s);
>          }
>          if (s->default_nan_mode) {
>              nan =  float16_default_nan(s);
> @@ -11978,7 +11978,7 @@ float32 HELPER(rsqrte_f32)(float32 input, void *fpstp)
>          float32 nan = f32;
>          if (float32_is_signaling_nan(f32, s)) {
>              float_raise(float_flag_invalid, s);
> -            nan = float32_maybe_silence_nan(f32, s);
> +            nan = float32_silence_nan(f32, s);
>          }
>          if (s->default_nan_mode) {
>              nan =  float32_default_nan(s);
> @@ -12021,7 +12021,7 @@ float64 HELPER(rsqrte_f64)(float64 input, void *fpstp)
>          float64 nan = f64;
>          if (float64_is_signaling_nan(f64, s)) {
>              float_raise(float_flag_invalid, s);
> -            nan = float64_maybe_silence_nan(f64, s);
> +            nan = float64_silence_nan(f64, s);
>          }
>          if (s->default_nan_mode) {
>              nan =  float64_default_nan(s);


--
Alex Bennée

[Qemu-devel] [PATCH v5 10/28] target/arm: convert conversion helpers to fpst/ahp_flag

[Richard Henderson]

From: Alex Bennée <alex.bennee@linaro.org>

Instead of passing env and leaving it up to the helper to get the
right fpstatus we pass it explicitly. There was already a get_fpstatus
helper for neon for the 32 bit code. We also add an get_ahp_flag() for
passing the state of the alternative FP16 format flag. This leaves
scope for later tracking the AHP state in translation flags.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
v4
  - remove neon_fcvt_*; they are now identical to vfp_fcvt_*.
  - add flags to vfp_fcvt_* helper decls.
  - add some missing tcg_temp_free_*.
v5
  - always use get_fpstatus_ptr(false) for FZ, since FZ16 is
    supposed to be supressed.
---
 target/arm/helper.h        | 10 +++---
 target/arm/translate.h     | 12 +++++++
 target/arm/helper.c        | 56 +++++------------------------
 target/arm/translate-a64.c | 38 ++++++++++++++++----
 target/arm/translate.c     | 74 +++++++++++++++++++++++++++++---------
 5 files changed, 113 insertions(+), 77 deletions(-)

diff --git a/target/arm/helper.h b/target/arm/helper.h
index ce89968b2d..047f3bc1ca 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -187,12 +187,10 @@ DEF_HELPER_3(vfp_uqtoh, f16, i64, i32, ptr)
 DEF_HELPER_FLAGS_2(set_rmode, TCG_CALL_NO_RWG, i32, i32, ptr)
 DEF_HELPER_FLAGS_2(set_neon_rmode, TCG_CALL_NO_RWG, i32, i32, env)
 
-DEF_HELPER_2(vfp_fcvt_f16_to_f32, f32, i32, env)
-DEF_HELPER_2(vfp_fcvt_f32_to_f16, i32, f32, env)
-DEF_HELPER_2(neon_fcvt_f16_to_f32, f32, i32, env)
-DEF_HELPER_2(neon_fcvt_f32_to_f16, i32, f32, env)
-DEF_HELPER_FLAGS_2(vfp_fcvt_f16_to_f64, TCG_CALL_NO_RWG, f64, i32, env)
-DEF_HELPER_FLAGS_2(vfp_fcvt_f64_to_f16, TCG_CALL_NO_RWG, i32, f64, env)
+DEF_HELPER_FLAGS_3(vfp_fcvt_f16_to_f32, TCG_CALL_NO_RWG, f32, f16, ptr, i32)
+DEF_HELPER_FLAGS_3(vfp_fcvt_f32_to_f16, TCG_CALL_NO_RWG, f16, f32, ptr, i32)
+DEF_HELPER_FLAGS_3(vfp_fcvt_f16_to_f64, TCG_CALL_NO_RWG, f64, f16, ptr, i32)
+DEF_HELPER_FLAGS_3(vfp_fcvt_f64_to_f16, TCG_CALL_NO_RWG, f16, f64, ptr, i32)
 
 DEF_HELPER_4(vfp_muladdd, f64, f64, f64, f64, ptr)
 DEF_HELPER_4(vfp_muladds, f32, f32, f32, f32, ptr)
diff --git a/target/arm/translate.h b/target/arm/translate.h
index 37a1bba056..45f04244be 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -177,4 +177,16 @@ void arm_free_cc(DisasCompare *cmp);
 void arm_jump_cc(DisasCompare *cmp, TCGLabel *label);
 void arm_gen_test_cc(int cc, TCGLabel *label);
 
+/* Return state of Alternate Half-precision flag, caller frees result */
+static inline TCGv_i32 get_ahp_flag(void)
+{
+    TCGv_i32 ret = tcg_temp_new_i32();
+
+    tcg_gen_ld_i32(ret, cpu_env,
+                   offsetof(CPUARMState, vfp.xregs[ARM_VFP_FPSCR]));
+    tcg_gen_extract_i32(ret, ret, 26, 1);
+
+    return ret;
+}
+
 #endif /* TARGET_ARM_TRANSLATE_H */
diff --git a/target/arm/helper.c b/target/arm/helper.c
index 3df5cf30e4..a1c1dc5bbe 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -11540,64 +11540,24 @@ uint32_t HELPER(set_neon_rmode)(uint32_t rmode, CPUARMState *env)
 }
 
 /* Half precision conversions.  */
-static float32 do_fcvt_f16_to_f32(uint32_t a, CPUARMState *env, float_status *s)
+float32 HELPER(vfp_fcvt_f16_to_f32)(float16 a, void *fpstp, uint32_t ahp_mode)
 {
-    int ieee = (env->vfp.xregs[ARM_VFP_FPSCR] & (1 << 26)) == 0;
-    float32 r = float16_to_float32(make_float16(a), ieee, s);
-    if (ieee) {
-        return float32_maybe_silence_nan(r, s);
-    }
-    return r;
+    return float16_to_float32(a, !ahp_mode, fpstp);
 }
 
-static uint32_t do_fcvt_f32_to_f16(float32 a, CPUARMState *env, float_status *s)
+float16 HELPER(vfp_fcvt_f32_to_f16)(float32 a, void *fpstp, uint32_t ahp_mode)
 {
-    int ieee = (env->vfp.xregs[ARM_VFP_FPSCR] & (1 << 26)) == 0;
-    float16 r = float32_to_float16(a, ieee, s);
-    if (ieee) {
-        r = float16_maybe_silence_nan(r, s);
-    }
-    return float16_val(r);
+    return float32_to_float16(a, !ahp_mode, fpstp);
 }
 
-float32 HELPER(neon_fcvt_f16_to_f32)(uint32_t a, CPUARMState *env)
+float64 HELPER(vfp_fcvt_f16_to_f64)(float16 a, void *fpstp, uint32_t ahp_mode)
 {
-    return do_fcvt_f16_to_f32(a, env, &env->vfp.standard_fp_status);
+    return float16_to_float64(a, !ahp_mode, fpstp);
 }
 
-uint32_t HELPER(neon_fcvt_f32_to_f16)(float32 a, CPUARMState *env)
+float16 HELPER(vfp_fcvt_f64_to_f16)(float64 a, void *fpstp, uint32_t ahp_mode)
 {
-    return do_fcvt_f32_to_f16(a, env, &env->vfp.standard_fp_status);
-}
-
-float32 HELPER(vfp_fcvt_f16_to_f32)(uint32_t a, CPUARMState *env)
-{
-    return do_fcvt_f16_to_f32(a, env, &env->vfp.fp_status);
-}
-
-uint32_t HELPER(vfp_fcvt_f32_to_f16)(float32 a, CPUARMState *env)
-{
-    return do_fcvt_f32_to_f16(a, env, &env->vfp.fp_status);
-}
-
-float64 HELPER(vfp_fcvt_f16_to_f64)(uint32_t a, CPUARMState *env)
-{
-    int ieee = (env->vfp.xregs[ARM_VFP_FPSCR] & (1 << 26)) == 0;
-    float64 r = float16_to_float64(make_float16(a), ieee, &env->vfp.fp_status);
-    if (ieee) {
-        return float64_maybe_silence_nan(r, &env->vfp.fp_status);
-    }
-    return r;
-}
-
-uint32_t HELPER(vfp_fcvt_f64_to_f16)(float64 a, CPUARMState *env)
-{
-    int ieee = (env->vfp.xregs[ARM_VFP_FPSCR] & (1 << 26)) == 0;
-    float16 r = float64_to_float16(a, ieee, &env->vfp.fp_status);
-    if (ieee) {
-        r = float16_maybe_silence_nan(r, &env->vfp.fp_status);
-    }
-    return float16_val(r);
+    return float64_to_float16(a, !ahp_mode, fpstp);
 }
 
 #define float32_two make_float32(0x40000000)
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index a0b0c43d12..4db18031bb 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -5147,10 +5147,15 @@ static void handle_fp_fcvt(DisasContext *s, int opcode,
         } else {
             /* Single to half */
             TCGv_i32 tcg_rd = tcg_temp_new_i32();
-            gen_helper_vfp_fcvt_f32_to_f16(tcg_rd, tcg_rn, cpu_env);
+            TCGv_i32 ahp = get_ahp_flag();
+            TCGv_ptr fpst = get_fpstatus_ptr(false);
+
+            gen_helper_vfp_fcvt_f32_to_f16(tcg_rd, tcg_rn, fpst, ahp);
             /* write_fp_sreg is OK here because top half of tcg_rd is zero */
             write_fp_sreg(s, rd, tcg_rd);
             tcg_temp_free_i32(tcg_rd);
+            tcg_temp_free_i32(ahp);
+            tcg_temp_free_ptr(fpst);
         }
         tcg_temp_free_i32(tcg_rn);
         break;
@@ -5163,9 +5168,13 @@ static void handle_fp_fcvt(DisasContext *s, int opcode,
             /* Double to single */
             gen_helper_vfp_fcvtsd(tcg_rd, tcg_rn, cpu_env);
         } else {
+            TCGv_ptr fpst = get_fpstatus_ptr(false);
+            TCGv_i32 ahp = get_ahp_flag();
             /* Double to half */
-            gen_helper_vfp_fcvt_f64_to_f16(tcg_rd, tcg_rn, cpu_env);
+            gen_helper_vfp_fcvt_f64_to_f16(tcg_rd, tcg_rn, fpst, ahp);
             /* write_fp_sreg is OK here because top half of tcg_rd is zero */
+            tcg_temp_free_ptr(fpst);
+            tcg_temp_free_i32(ahp);
         }
         write_fp_sreg(s, rd, tcg_rd);
         tcg_temp_free_i32(tcg_rd);
@@ -5175,17 +5184,21 @@ static void handle_fp_fcvt(DisasContext *s, int opcode,
     case 0x3:
     {
         TCGv_i32 tcg_rn = read_fp_sreg(s, rn);
+        TCGv_ptr tcg_fpst = get_fpstatus_ptr(false);
+        TCGv_i32 tcg_ahp = get_ahp_flag();
         tcg_gen_ext16u_i32(tcg_rn, tcg_rn);
         if (dtype == 0) {
             /* Half to single */
             TCGv_i32 tcg_rd = tcg_temp_new_i32();
-            gen_helper_vfp_fcvt_f16_to_f32(tcg_rd, tcg_rn, cpu_env);
+            gen_helper_vfp_fcvt_f16_to_f32(tcg_rd, tcg_rn, tcg_fpst, tcg_ahp);
             write_fp_sreg(s, rd, tcg_rd);
+            tcg_temp_free_ptr(tcg_fpst);
+            tcg_temp_free_i32(tcg_ahp);
             tcg_temp_free_i32(tcg_rd);
         } else {
             /* Half to double */
             TCGv_i64 tcg_rd = tcg_temp_new_i64();
-            gen_helper_vfp_fcvt_f16_to_f64(tcg_rd, tcg_rn, cpu_env);
+            gen_helper_vfp_fcvt_f16_to_f64(tcg_rd, tcg_rn, tcg_fpst, tcg_ahp);
             write_fp_dreg(s, rd, tcg_rd);
             tcg_temp_free_i64(tcg_rd);
         }
@@ -9053,12 +9066,17 @@ static void handle_2misc_narrow(DisasContext *s, bool scalar,
             } else {
                 TCGv_i32 tcg_lo = tcg_temp_new_i32();
                 TCGv_i32 tcg_hi = tcg_temp_new_i32();
+                TCGv_ptr fpst = get_fpstatus_ptr(true);
+                TCGv_i32 ahp = get_ahp_flag();
+
                 tcg_gen_extr_i64_i32(tcg_lo, tcg_hi, tcg_op);
-                gen_helper_vfp_fcvt_f32_to_f16(tcg_lo, tcg_lo, cpu_env);
-                gen_helper_vfp_fcvt_f32_to_f16(tcg_hi, tcg_hi, cpu_env);
+                gen_helper_vfp_fcvt_f32_to_f16(tcg_lo, tcg_lo, fpst, ahp);
+                gen_helper_vfp_fcvt_f32_to_f16(tcg_hi, tcg_hi, fpst, ahp);
                 tcg_gen_deposit_i32(tcg_res[pass], tcg_lo, tcg_hi, 16, 16);
                 tcg_temp_free_i32(tcg_lo);
                 tcg_temp_free_i32(tcg_hi);
+                tcg_temp_free_ptr(fpst);
+                tcg_temp_free_i32(ahp);
             }
             break;
         case 0x56:  /* FCVTXN, FCVTXN2 */
@@ -11532,18 +11550,24 @@ static void handle_2misc_widening(DisasContext *s, int opcode, bool is_q,
         /* 16 -> 32 bit fp conversion */
         int srcelt = is_q ? 4 : 0;
         TCGv_i32 tcg_res[4];
+        TCGv_ptr fpst = get_fpstatus_ptr(true);
+        TCGv_i32 ahp = get_ahp_flag();
+
 
         for (pass = 0; pass < 4; pass++) {
             tcg_res[pass] = tcg_temp_new_i32();
 
             read_vec_element_i32(s, tcg_res[pass], rn, srcelt + pass, MO_16);
             gen_helper_vfp_fcvt_f16_to_f32(tcg_res[pass], tcg_res[pass],
-                                           cpu_env);
+                                           fpst, ahp);
         }
         for (pass = 0; pass < 4; pass++) {
             write_vec_element_i32(s, tcg_res[pass], rd, pass, MO_32);
             tcg_temp_free_i32(tcg_res[pass]);
         }
+
+        tcg_temp_free_ptr(fpst);
+        tcg_temp_free_i32(ahp);
     }
 }
 
diff --git a/target/arm/translate.c b/target/arm/translate.c
index 731cf327a1..5ba59e0188 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -3824,38 +3824,56 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         gen_vfp_sqrt(dp);
                         break;
                     case 4: /* vcvtb.f32.f16, vcvtb.f64.f16 */
+                    {
+                        TCGv_ptr fpst = get_fpstatus_ptr(false);
+                        TCGv_i32 ahp_mode = get_ahp_flag();
                         tmp = gen_vfp_mrs();
                         tcg_gen_ext16u_i32(tmp, tmp);
                         if (dp) {
                             gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
-                                                           cpu_env);
+                                                           fpst, ahp_mode);
                         } else {
                             gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
-                                                           cpu_env);
+                                                           fpst, ahp_mode);
                         }
+                        tcg_temp_free_i32(ahp_mode);
+                        tcg_temp_free_ptr(fpst);
                         tcg_temp_free_i32(tmp);
                         break;
+                    }
                     case 5: /* vcvtt.f32.f16, vcvtt.f64.f16 */
+                    {
+                        TCGv_ptr fpst = get_fpstatus_ptr(false);
+                        TCGv_i32 ahp = get_ahp_flag();
                         tmp = gen_vfp_mrs();
                         tcg_gen_shri_i32(tmp, tmp, 16);
                         if (dp) {
                             gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
-                                                           cpu_env);
+                                                           fpst, ahp);
                         } else {
                             gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
-                                                           cpu_env);
+                                                           fpst, ahp);
                         }
                         tcg_temp_free_i32(tmp);
+                        tcg_temp_free_i32(ahp);
+                        tcg_temp_free_ptr(fpst);
                         break;
+                    }
                     case 6: /* vcvtb.f16.f32, vcvtb.f16.f64 */
+                    {
+                        TCGv_ptr fpst = get_fpstatus_ptr(false);
+                        TCGv_i32 ahp = get_ahp_flag();
                         tmp = tcg_temp_new_i32();
+
                         if (dp) {
                             gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
-                                                           cpu_env);
+                                                           fpst, ahp);
                         } else {
                             gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
-                                                           cpu_env);
+                                                           fpst, ahp);
                         }
+                        tcg_temp_free_i32(ahp);
+                        tcg_temp_free_ptr(fpst);
                         gen_mov_F0_vreg(0, rd);
                         tmp2 = gen_vfp_mrs();
                         tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
@@ -3863,15 +3881,21 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         tcg_temp_free_i32(tmp2);
                         gen_vfp_msr(tmp);
                         break;
+                    }
                     case 7: /* vcvtt.f16.f32, vcvtt.f16.f64 */
+                    {
+                        TCGv_ptr fpst = get_fpstatus_ptr(true);
+                        TCGv_i32 ahp = get_ahp_flag();
                         tmp = tcg_temp_new_i32();
                         if (dp) {
                             gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
-                                                           cpu_env);
+                                                           fpst, ahp);
                         } else {
                             gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
-                                                           cpu_env);
+                                                           fpst, ahp);
                         }
+                        tcg_temp_free_i32(ahp);
+                        tcg_temp_free_ptr(fpst);
                         tcg_gen_shli_i32(tmp, tmp, 16);
                         gen_mov_F0_vreg(0, rd);
                         tmp2 = gen_vfp_mrs();
@@ -3880,6 +3904,7 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         tcg_temp_free_i32(tmp2);
                         gen_vfp_msr(tmp);
                         break;
+                    }
                     case 8: /* cmp */
                         gen_vfp_cmp(dp);
                         break;
@@ -7222,53 +7247,70 @@ static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
                     }
                     break;
                 case NEON_2RM_VCVT_F16_F32:
+                {
+                    TCGv_ptr fpst;
+                    TCGv_i32 ahp;
+
                     if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
                         q || (rm & 1)) {
                         return 1;
                     }
                     tmp = tcg_temp_new_i32();
                     tmp2 = tcg_temp_new_i32();
+                    fpst = get_fpstatus_ptr(true);
+                    ahp = get_ahp_flag();
                     tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 0));
-                    gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
+                    gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s, fpst, ahp);
                     tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 1));
-                    gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
+                    gen_helper_vfp_fcvt_f32_to_f16(tmp2, cpu_F0s, fpst, ahp);
                     tcg_gen_shli_i32(tmp2, tmp2, 16);
                     tcg_gen_or_i32(tmp2, tmp2, tmp);
                     tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 2));
-                    gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
+                    gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s, fpst, ahp);
                     tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 3));
                     neon_store_reg(rd, 0, tmp2);
                     tmp2 = tcg_temp_new_i32();
-                    gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
+                    gen_helper_vfp_fcvt_f32_to_f16(tmp2, cpu_F0s, fpst, ahp);
                     tcg_gen_shli_i32(tmp2, tmp2, 16);
                     tcg_gen_or_i32(tmp2, tmp2, tmp);
                     neon_store_reg(rd, 1, tmp2);
                     tcg_temp_free_i32(tmp);
+                    tcg_temp_free_i32(ahp);
+                    tcg_temp_free_ptr(fpst);
                     break;
+                }
                 case NEON_2RM_VCVT_F32_F16:
+                {
+                    TCGv_ptr fpst;
+                    TCGv_i32 ahp;
                     if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
                         q || (rd & 1)) {
                         return 1;
                     }
+                    fpst = get_fpstatus_ptr(true);
+                    ahp = get_ahp_flag();
                     tmp3 = tcg_temp_new_i32();
                     tmp = neon_load_reg(rm, 0);
                     tmp2 = neon_load_reg(rm, 1);
                     tcg_gen_ext16u_i32(tmp3, tmp);
-                    gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
+                    gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
                     tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 0));
                     tcg_gen_shri_i32(tmp3, tmp, 16);
-                    gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
+                    gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
                     tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 1));
                     tcg_temp_free_i32(tmp);
                     tcg_gen_ext16u_i32(tmp3, tmp2);
-                    gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
+                    gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
                     tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 2));
                     tcg_gen_shri_i32(tmp3, tmp2, 16);
-                    gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
+                    gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp3, fpst, ahp);
                     tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 3));
                     tcg_temp_free_i32(tmp2);
                     tcg_temp_free_i32(tmp3);
+                    tcg_temp_free_i32(ahp);
+                    tcg_temp_free_ptr(fpst);
                     break;
+                }
                 case NEON_2RM_AESE: case NEON_2RM_AESMC:
                     if (!arm_dc_feature(s, ARM_FEATURE_V8_AES)
                         || ((rm | rd) & 1)) {
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 10/28] target/arm: convert conversion helpers to fpst/ahp_flag

[Peter Maydell]

On 14 May 2018 at 23:12, Richard Henderson <richard.henderson@linaro.org> wrote:
> From: Alex Bennée <alex.bennee@linaro.org>
>
> Instead of passing env and leaving it up to the helper to get the
> right fpstatus we pass it explicitly. There was already a get_fpstatus
> helper for neon for the 32 bit code. We also add an get_ahp_flag() for
> passing the state of the alternative FP16 format flag. This leaves
> scope for later tracking the AHP state in translation flags.
>
> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>
> ---
> v4
>   - remove neon_fcvt_*; they are now identical to vfp_fcvt_*.
>   - add flags to vfp_fcvt_* helper decls.
>   - add some missing tcg_temp_free_*.
> v5
>   - always use get_fpstatus_ptr(false) for FZ, since FZ16 is
>     supposed to be supressed.
> ---

> @@ -9053,12 +9066,17 @@ static void handle_2misc_narrow(DisasContext *s, bool scalar,
>              } else {
>                  TCGv_i32 tcg_lo = tcg_temp_new_i32();
>                  TCGv_i32 tcg_hi = tcg_temp_new_i32();
> +                TCGv_ptr fpst = get_fpstatus_ptr(true);

Still the wrong fpstatus here...

> +                TCGv_i32 ahp = get_ahp_flag();
> +
>                  tcg_gen_extr_i64_i32(tcg_lo, tcg_hi, tcg_op);
> -                gen_helper_vfp_fcvt_f32_to_f16(tcg_lo, tcg_lo, cpu_env);
> -                gen_helper_vfp_fcvt_f32_to_f16(tcg_hi, tcg_hi, cpu_env);
> +                gen_helper_vfp_fcvt_f32_to_f16(tcg_lo, tcg_lo, fpst, ahp);
> +                gen_helper_vfp_fcvt_f32_to_f16(tcg_hi, tcg_hi, fpst, ahp);
>                  tcg_gen_deposit_i32(tcg_res[pass], tcg_lo, tcg_hi, 16, 16);
>                  tcg_temp_free_i32(tcg_lo);
>                  tcg_temp_free_i32(tcg_hi);
> +                tcg_temp_free_ptr(fpst);
> +                tcg_temp_free_i32(ahp);
>              }
>              break;
>          case 0x56:  /* FCVTXN, FCVTXN2 */

> @@ -11532,18 +11550,24 @@ static void handle_2misc_widening(DisasContext *s, int opcode, bool is_q,
>          /* 16 -> 32 bit fp conversion */
>          int srcelt = is_q ? 4 : 0;
>          TCGv_i32 tcg_res[4];
> +        TCGv_ptr fpst = get_fpstatus_ptr(true);

...and here.

> +        TCGv_i32 ahp = get_ahp_flag();
> +
>
>          for (pass = 0; pass < 4; pass++) {
>              tcg_res[pass] = tcg_temp_new_i32();
>
>              read_vec_element_i32(s, tcg_res[pass], rn, srcelt + pass, MO_16);
>              gen_helper_vfp_fcvt_f16_to_f32(tcg_res[pass], tcg_res[pass],
> -                                           cpu_env);
> +                                           fpst, ahp);
>          }
>          for (pass = 0; pass < 4; pass++) {
>              write_vec_element_i32(s, tcg_res[pass], rd, pass, MO_32);
>              tcg_temp_free_i32(tcg_res[pass]);
>          }
> +
> +        tcg_temp_free_ptr(fpst);
> +        tcg_temp_free_i32(ahp);
>      }
>  }

thanks
-- PMM

Re: [Qemu-devel] [PATCH v5 10/28] target/arm: convert conversion helpers to fpst/ahp_flag

[Richard Henderson]

On 05/15/2018 03:08 AM, Peter Maydell wrote:
> On 14 May 2018 at 23:12, Richard Henderson <richard.henderson@linaro.org> wrote:
>> From: Alex Bennée <alex.bennee@linaro.org>
>>
>> Instead of passing env and leaving it up to the helper to get the
>> right fpstatus we pass it explicitly. There was already a get_fpstatus
>> helper for neon for the 32 bit code. We also add an get_ahp_flag() for
>> passing the state of the alternative FP16 format flag. This leaves
>> scope for later tracking the AHP state in translation flags.
>>
>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>>
>> ---
>> v4
>>   - remove neon_fcvt_*; they are now identical to vfp_fcvt_*.
>>   - add flags to vfp_fcvt_* helper decls.
>>   - add some missing tcg_temp_free_*.
>> v5
>>   - always use get_fpstatus_ptr(false) for FZ, since FZ16 is
>>     supposed to be supressed.
>> ---
> 
>> @@ -9053,12 +9066,17 @@ static void handle_2misc_narrow(DisasContext *s, bool scalar,
>>              } else {
>>                  TCGv_i32 tcg_lo = tcg_temp_new_i32();
>>                  TCGv_i32 tcg_hi = tcg_temp_new_i32();
>> +                TCGv_ptr fpst = get_fpstatus_ptr(true);
> 
> Still the wrong fpstatus here...

Bah.  Got em now.  Thanks.


r~

[Qemu-devel] [PATCH v5 11/28] target/arm: squash FZ16 behaviour for conversions

[Richard Henderson]

From: Alex Bennée <alex.bennee@linaro.org>

The ARM ARM specifies FZ16 is suppressed for conversions. Rather than
pushing this logic into the softfloat code we can simply save the FZ
state and temporarily disable it for the softfloat call.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
v4
  - float16_to_floatX squished the wrong softfloat bit for FZ16;
    need to adjust input denormals in this case.
---
 target/arm/helper.c | 40 ++++++++++++++++++++++++++++++++++++----
 1 file changed, 36 insertions(+), 4 deletions(-)

diff --git a/target/arm/helper.c b/target/arm/helper.c
index a1c1dc5bbe..e05c7230d4 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -11542,22 +11542,54 @@ uint32_t HELPER(set_neon_rmode)(uint32_t rmode, CPUARMState *env)
 /* Half precision conversions.  */
 float32 HELPER(vfp_fcvt_f16_to_f32)(float16 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float16_to_float32(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing input denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_inputs_to_zero(fpst);
+    set_flush_inputs_to_zero(false, fpst);
+    float32 r = float16_to_float32(a, !ahp_mode, fpst);
+    set_flush_inputs_to_zero(save, fpst);
+    return r;
 }
 
 float16 HELPER(vfp_fcvt_f32_to_f16)(float32 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float32_to_float16(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing output denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_to_zero(fpst);
+    set_flush_to_zero(false, fpst);
+    float16 r = float32_to_float16(a, !ahp_mode, fpst);
+    set_flush_to_zero(save, fpst);
+    return r;
 }
 
 float64 HELPER(vfp_fcvt_f16_to_f64)(float16 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float16_to_float64(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing input denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_inputs_to_zero(fpst);
+    set_flush_inputs_to_zero(false, fpst);
+    float64 r = float16_to_float64(a, !ahp_mode, fpst);
+    set_flush_inputs_to_zero(save, fpst);
+    return r;
 }
 
 float16 HELPER(vfp_fcvt_f64_to_f16)(float64 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float64_to_float16(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing output denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_to_zero(fpst);
+    set_flush_to_zero(false, fpst);
+    float16 r = float64_to_float16(a, !ahp_mode, fpst);
+    set_flush_to_zero(save, fpst);
+    return r;
 }
 
 #define float32_two make_float32(0x40000000)
-- 
2.17.0

[Qemu-devel] [PATCH v5 12/28] target/arm: Remove floatX_maybe_silence_nan from conversions

[Richard Henderson]

This is now handled properly by the generic softfloat code.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/helper-a64.c |  1 -
 target/arm/helper.c     | 12 ++----------
 2 files changed, 2 insertions(+), 11 deletions(-)

diff --git a/target/arm/helper-a64.c b/target/arm/helper-a64.c
index 6f0eb83661..f92bdea732 100644
--- a/target/arm/helper-a64.c
+++ b/target/arm/helper-a64.c
@@ -466,7 +466,6 @@ float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
     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, &tstat);
     exflags = get_float_exception_flags(&tstat);
     if (exflags & float_flag_inexact) {
         r = make_float32(float32_val(r) | 1);
diff --git a/target/arm/helper.c b/target/arm/helper.c
index e05c7230d4..db8bbe52a6 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -11348,20 +11348,12 @@ FLOAT_CONVS(ui, d, 64, u)
 /* floating point conversion */
 float64 VFP_HELPER(fcvtd, s)(float32 x, CPUARMState *env)
 {
-    float64 r = float32_to_float64(x, &env->vfp.fp_status);
-    /* ARM requires that S<->D conversion of any kind of NaN generates
-     * a quiet NaN by forcing the most significant frac bit to 1.
-     */
-    return float64_maybe_silence_nan(r, &env->vfp.fp_status);
+    return float32_to_float64(x, &env->vfp.fp_status);
 }
 
 float32 VFP_HELPER(fcvts, d)(float64 x, CPUARMState *env)
 {
-    float32 r =  float64_to_float32(x, &env->vfp.fp_status);
-    /* ARM requires that S<->D conversion of any kind of NaN generates
-     * a quiet NaN by forcing the most significant frac bit to 1.
-     */
-    return float32_maybe_silence_nan(r, &env->vfp.fp_status);
+    return float64_to_float32(x, &env->vfp.fp_status);
 }
 
 /* VFP3 fixed point conversion.  */
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 12/28] target/arm: Remove floatX_maybe_silence_nan from conversions

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> This is now handled properly by the generic softfloat code.
>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  target/arm/helper-a64.c |  1 -
>  target/arm/helper.c     | 12 ++----------
>  2 files changed, 2 insertions(+), 11 deletions(-)
>
> diff --git a/target/arm/helper-a64.c b/target/arm/helper-a64.c
> index 6f0eb83661..f92bdea732 100644
> --- a/target/arm/helper-a64.c
> +++ b/target/arm/helper-a64.c
> @@ -466,7 +466,6 @@ float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
>      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, &tstat);
>      exflags = get_float_exception_flags(&tstat);
>      if (exflags & float_flag_inexact) {
>          r = make_float32(float32_val(r) | 1);
> diff --git a/target/arm/helper.c b/target/arm/helper.c
> index e05c7230d4..db8bbe52a6 100644
> --- a/target/arm/helper.c
> +++ b/target/arm/helper.c
> @@ -11348,20 +11348,12 @@ FLOAT_CONVS(ui, d, 64, u)
>  /* floating point conversion */
>  float64 VFP_HELPER(fcvtd, s)(float32 x, CPUARMState *env)
>  {
> -    float64 r = float32_to_float64(x, &env->vfp.fp_status);
> -    /* ARM requires that S<->D conversion of any kind of NaN generates
> -     * a quiet NaN by forcing the most significant frac bit to 1.
> -     */
> -    return float64_maybe_silence_nan(r, &env->vfp.fp_status);
> +    return float32_to_float64(x, &env->vfp.fp_status);
>  }
>
>  float32 VFP_HELPER(fcvts, d)(float64 x, CPUARMState *env)
>  {
> -    float32 r =  float64_to_float32(x, &env->vfp.fp_status);
> -    /* ARM requires that S<->D conversion of any kind of NaN generates
> -     * a quiet NaN by forcing the most significant frac bit to 1.
> -     */
> -    return float32_maybe_silence_nan(r, &env->vfp.fp_status);
> +    return float64_to_float32(x, &env->vfp.fp_status);
>  }
>
>  /* VFP3 fixed point conversion.  */


--
Alex Bennée

[Qemu-devel] [PATCH v5 13/28] fpu/softfloat: Partial support for ARM Alternative half-precision

[Richard Henderson]

From: Alex Bennée <alex.bennee@linaro.org>

For float16 ARM supports an alternative half-precision format which
sacrifices the ability to represent NaN/Inf in return for a higher
dynamic range.  The new FloatFmt flag, arm_althp, is then used to
modify the behaviour of canonicalize and round_canonical with respect
to representation and exception raising.

Usage of this new flag waits until we re-factor float-to-float conversions.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
v3
  - squash NaN to 0 if destination is AHP F16
v4
  - handle inf -> ahp max in float_to_float not round_canonical
  - assert no nan and inf for ahp in round_canonical
  - check ahp before snan in float_to_float
v5
  - split out canonicalize and round_canonical changes from the rest
---
 fpu/softfloat.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 41253c6749..55d0d01ec3 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -220,8 +220,10 @@ typedef struct {
  *   frac_shift: shift to normalise the fraction with DECOMPOSED_BINARY_POINT
  * The following are computed based the size of fraction
  *   frac_lsb: least significant bit of fraction
- *   fram_lsbm1: the bit bellow the least significant bit (for rounding)
+ *   frac_lsbm1: the bit bellow the least significant bit (for rounding)
  *   round_mask/roundeven_mask: masks used for rounding
+ * The following optional modifiers are available:
+ *   arm_althp: handle ARM Alternative Half Precision
  */
 typedef struct {
     int exp_size;
@@ -233,6 +235,7 @@ typedef struct {
     uint64_t frac_lsbm1;
     uint64_t round_mask;
     uint64_t roundeven_mask;
+    bool arm_althp;
 } FloatFmt;
 
 /* Expand fields based on the size of exponent and fraction */
@@ -324,7 +327,7 @@ static inline float64 float64_pack_raw(FloatParts p)
 static FloatParts canonicalize(FloatParts part, const FloatFmt *parm,
                                float_status *status)
 {
-    if (part.exp == parm->exp_max) {
+    if (part.exp == parm->exp_max && !parm->arm_althp) {
         if (part.frac == 0) {
             part.cls = float_class_inf;
         } else {
@@ -413,7 +416,15 @@ static FloatParts round_canonical(FloatParts p, float_status *s,
             }
             frac >>= frac_shift;
 
-            if (unlikely(exp >= exp_max)) {
+            if (parm->arm_althp) {
+                /* ARM Alt HP eschews Inf and NaN for a wider exponent.  */
+                if (unlikely(exp > exp_max)) {
+                    /* Overflow.  Return the maximum normal.  */
+                    flags = float_flag_invalid;
+                    exp = exp_max;
+                    frac = -1;
+                }
+            } else if (unlikely(exp >= exp_max)) {
                 flags |= float_flag_overflow | float_flag_inexact;
                 if (overflow_norm) {
                     exp = exp_max - 1;
@@ -464,12 +475,14 @@ static FloatParts round_canonical(FloatParts p, float_status *s,
 
     case float_class_inf:
     do_inf:
+        assert(!parm->arm_althp);
         exp = exp_max;
         frac = 0;
         break;
 
     case float_class_qnan:
     case float_class_snan:
+        assert(!parm->arm_althp);
         exp = exp_max;
         frac >>= parm->frac_shift;
         break;
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 13/28] fpu/softfloat: Partial support for ARM Alternative half-precision

[Peter Maydell]

On 14 May 2018 at 23:12, Richard Henderson <richard.henderson@linaro.org> wrote:
> From: Alex Bennée <alex.bennee@linaro.org>
>
> For float16 ARM supports an alternative half-precision format which
> sacrifices the ability to represent NaN/Inf in return for a higher
> dynamic range.  The new FloatFmt flag, arm_althp, is then used to
> modify the behaviour of canonicalize and round_canonical with respect
> to representation and exception raising.
>
> Usage of this new flag waits until we re-factor float-to-float conversions.
>
> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>
> ---
> v3
>   - squash NaN to 0 if destination is AHP F16
> v4
>   - handle inf -> ahp max in float_to_float not round_canonical
>   - assert no nan and inf for ahp in round_canonical
>   - check ahp before snan in float_to_float
> v5
>   - split out canonicalize and round_canonical changes from the rest
> ---
>  fpu/softfloat.c | 19 ++++++++++++++++---
>  1 file changed, 16 insertions(+), 3 deletions(-)
>
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index 41253c6749..55d0d01ec3 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -220,8 +220,10 @@ typedef struct {
>   *   frac_shift: shift to normalise the fraction with DECOMPOSED_BINARY_POINT
>   * The following are computed based the size of fraction
>   *   frac_lsb: least significant bit of fraction
> - *   fram_lsbm1: the bit bellow the least significant bit (for rounding)
> + *   frac_lsbm1: the bit bellow the least significant bit (for rounding)

"below"

>   *   round_mask/roundeven_mask: masks used for rounding
> + * The following optional modifiers are available:
> + *   arm_althp: handle ARM Alternative Half Precision
>   */

Otherwise
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>

thanks
-- PMM

[Qemu-devel] [PATCH v5 14/28] fpu/softfloat: re-factor float to float conversions

[Richard Henderson]

From: Alex Bennée <alex.bennee@linaro.org>

This allows us to delete a lot of additional boilerplate
code which is no longer needed.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
v2
  - pass FloatFmt to float_to_float instead of sizes
  - split AHP handling to another patch
  - use rth's suggested re-packing (+ setting .exp)
v3
  - also rm extractFloat16Sign
v4
  - update for canonical nan handling
v5
  - merge arm alt fp16 support into this patch
---
 fpu/softfloat-specialize.h |  40 ---
 include/fpu/softfloat.h    |   8 +-
 fpu/softfloat.c            | 488 +++++++++----------------------------
 3 files changed, 122 insertions(+), 414 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 571d1df378..995a0132c6 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -377,46 +377,6 @@ float16 float16_maybe_silence_nan(float16 a, float_status *status)
     return a;
 }
 
-/*----------------------------------------------------------------------------
-| Returns the result of converting the half-precision floating-point NaN
-| `a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
-| exception is raised.
-*----------------------------------------------------------------------------*/
-
-static commonNaNT float16ToCommonNaN(float16 a, float_status *status)
-{
-    commonNaNT z;
-
-    if (float16_is_signaling_nan(a, status)) {
-        float_raise(float_flag_invalid, status);
-    }
-    z.sign = float16_val(a) >> 15;
-    z.low = 0;
-    z.high = ((uint64_t) float16_val(a)) << 54;
-    return z;
-}
-
-/*----------------------------------------------------------------------------
-| Returns the result of converting the canonical NaN `a' to the half-
-| precision floating-point format.
-*----------------------------------------------------------------------------*/
-
-static float16 commonNaNToFloat16(commonNaNT a, float_status *status)
-{
-    uint16_t mantissa = a.high >> 54;
-
-    if (status->default_nan_mode) {
-        return float16_default_nan(status);
-    }
-
-    if (mantissa) {
-        return make_float16(((((uint16_t) a.sign) << 15)
-                             | (0x1F << 10) | mantissa));
-    } else {
-        return float16_default_nan(status);
-    }
-}
-
 /*----------------------------------------------------------------------------
 | Returns 1 if the single-precision floating-point value `a' is a quiet
 | NaN; otherwise returns 0.
diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h
index 43962dc3f5..a6860e858d 100644
--- a/include/fpu/softfloat.h
+++ b/include/fpu/softfloat.h
@@ -211,10 +211,10 @@ float128 uint64_to_float128(uint64_t, float_status *status);
 /*----------------------------------------------------------------------------
 | Software half-precision conversion routines.
 *----------------------------------------------------------------------------*/
-float16 float32_to_float16(float32, flag, float_status *status);
-float32 float16_to_float32(float16, flag, float_status *status);
-float16 float64_to_float16(float64 a, flag ieee, float_status *status);
-float64 float16_to_float64(float16 a, flag ieee, float_status *status);
+float16 float32_to_float16(float32, bool ieee, float_status *status);
+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);
 int16_t float16_to_int16(float16, float_status *status);
 uint16_t float16_to_uint16(float16 a, float_status *status);
 int16_t float16_to_int16_round_to_zero(float16, float_status *status);
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 55d0d01ec3..e9e421728b 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -113,15 +113,6 @@ static inline int extractFloat16Exp(float16 a)
     return (float16_val(a) >> 10) & 0x1f;
 }
 
-/*----------------------------------------------------------------------------
-| Returns the sign bit of the single-precision floating-point value `a'.
-*----------------------------------------------------------------------------*/
-
-static inline flag extractFloat16Sign(float16 a)
-{
-    return float16_val(a)>>15;
-}
-
 /*----------------------------------------------------------------------------
 | Returns the fraction bits of the single-precision floating-point value `a'.
 *----------------------------------------------------------------------------*/
@@ -254,6 +245,11 @@ static const FloatFmt float16_params = {
     FLOAT_PARAMS(5, 10)
 };
 
+static const FloatFmt float16_params_ahp = {
+    FLOAT_PARAMS(5, 10),
+    .arm_althp = true
+};
+
 static const FloatFmt float32_params = {
     FLOAT_PARAMS(8, 23)
 };
@@ -497,14 +493,27 @@ static FloatParts round_canonical(FloatParts p, float_status *s,
     return p;
 }
 
+/* Explicit FloatFmt version */
+static FloatParts float16a_unpack_canonical(float16 f, float_status *s,
+                                            const FloatFmt *params)
+{
+    return canonicalize(float16_unpack_raw(f), params, s);
+}
+
 static FloatParts float16_unpack_canonical(float16 f, float_status *s)
 {
-    return canonicalize(float16_unpack_raw(f), &float16_params, s);
+    return float16a_unpack_canonical(f, s, &float16_params);
+}
+
+static float16 float16a_round_pack_canonical(FloatParts p, float_status *s,
+                                             const FloatFmt *params)
+{
+    return float16_pack_raw(round_canonical(p, s, params));
 }
 
 static float16 float16_round_pack_canonical(FloatParts p, float_status *s)
 {
-    return float16_pack_raw(round_canonical(p, s, &float16_params));
+    return float16a_round_pack_canonical(p, s, &float16_params);
 }
 
 static FloatParts float32_unpack_canonical(float32 f, float_status *s)
@@ -1181,6 +1190,104 @@ float64 float64_div(float64 a, float64 b, float_status *status)
     return float64_round_pack_canonical(pr, status);
 }
 
+/*
+ * Float to Float conversions
+ *
+ * Returns the result of converting one float format to another. The
+ * conversion is performed according to the IEC/IEEE Standard for
+ * Binary Floating-Point Arithmetic.
+ *
+ * The float_to_float helper only needs to take care of raising
+ * invalid exceptions and handling the conversion on NaNs.
+ */
+
+static FloatParts float_to_float(FloatParts a, const FloatFmt *dstf,
+                                 float_status *s)
+{
+    if (dstf->arm_althp) {
+        switch (a.cls) {
+        case float_class_qnan:
+        case float_class_snan:
+            /* There is no NaN in the destination format.  Raise Invalid
+             * and return a zero with the sign of the input NaN.
+             */
+            s->float_exception_flags |= float_flag_invalid;
+            a.cls = float_class_zero;
+            a.frac = 0;
+            a.exp = 0;
+            break;
+
+        case float_class_inf:
+            /* There is no Inf in the destination format.  Raise Invalid
+             * and return the maximum normal with the correct sign.
+             */
+            s->float_exception_flags |= float_flag_invalid;
+            a.cls = float_class_normal;
+            a.exp = dstf->exp_max;
+            a.frac = ((1ull << dstf->frac_size) - 1) << dstf->frac_shift;
+            break;
+
+        default:
+            break;
+        }
+    } else if (is_nan(a.cls)) {
+        if (is_snan(a.cls)) {
+            s->float_exception_flags |= float_flag_invalid;
+            a = parts_silence_nan(a, s);
+        }
+        if (s->default_nan_mode) {
+            return parts_default_nan(s);
+        }
+    }
+    return a;
+}
+
+float32 float16_to_float32(float16 a, bool ieee, float_status *s)
+{
+    const FloatFmt *fmt16 = ieee ? &float16_params : &float16_params_ahp;
+    FloatParts p = float16a_unpack_canonical(a, s, fmt16);
+    FloatParts pr = float_to_float(p, &float32_params, s);
+    return float32_round_pack_canonical(pr, s);
+}
+
+float64 float16_to_float64(float16 a, bool ieee, float_status *s)
+{
+    const FloatFmt *fmt16 = ieee ? &float16_params : &float16_params_ahp;
+    FloatParts p = float16a_unpack_canonical(a, s, fmt16);
+    FloatParts pr = float_to_float(p, &float64_params, s);
+    return float64_round_pack_canonical(pr, s);
+}
+
+float16 float32_to_float16(float32 a, bool ieee, float_status *s)
+{
+    const FloatFmt *fmt16 = ieee ? &float16_params : &float16_params_ahp;
+    FloatParts p = float32_unpack_canonical(a, s);
+    FloatParts pr = float_to_float(p, fmt16, s);
+    return float16a_round_pack_canonical(pr, s, fmt16);
+}
+
+float64 float32_to_float64(float32 a, float_status *s)
+{
+    FloatParts p = float32_unpack_canonical(a, s);
+    FloatParts pr = float_to_float(p, &float64_params, s);
+    return float64_round_pack_canonical(pr, s);
+}
+
+float16 float64_to_float16(float64 a, bool ieee, float_status *s)
+{
+    const FloatFmt *fmt16 = ieee ? &float16_params : &float16_params_ahp;
+    FloatParts p = float64_unpack_canonical(a, s);
+    FloatParts pr = float_to_float(p, fmt16, s);
+    return float16a_round_pack_canonical(pr, s, fmt16);
+}
+
+float32 float64_to_float32(float64 a, float_status *s)
+{
+    FloatParts p = float64_unpack_canonical(a, s);
+    FloatParts pr = float_to_float(p, &float32_params, s);
+    return float32_round_pack_canonical(pr, s);
+}
+
 /*
  * Rounds the floating-point value `a' to an integer, and returns the
  * result as a floating-point value. The operation is performed
@@ -3124,41 +3231,6 @@ float128 uint64_to_float128(uint64_t a, float_status *status)
     return normalizeRoundAndPackFloat128(0, 0x406E, 0, a, status);
 }
 
-
-
-
-/*----------------------------------------------------------------------------
-| Returns the result of converting the single-precision floating-point value
-| `a' to the double-precision floating-point format.  The conversion is
-| performed according to the IEC/IEEE Standard for Binary Floating-Point
-| Arithmetic.
-*----------------------------------------------------------------------------*/
-
-float64 float32_to_float64(float32 a, float_status *status)
-{
-    flag aSign;
-    int aExp;
-    uint32_t aSig;
-    a = float32_squash_input_denormal(a, status);
-
-    aSig = extractFloat32Frac( a );
-    aExp = extractFloat32Exp( a );
-    aSign = extractFloat32Sign( a );
-    if ( aExp == 0xFF ) {
-        if (aSig) {
-            return commonNaNToFloat64(float32ToCommonNaN(a, status), status);
-        }
-        return packFloat64( aSign, 0x7FF, 0 );
-    }
-    if ( aExp == 0 ) {
-        if ( aSig == 0 ) return packFloat64( aSign, 0, 0 );
-        normalizeFloat32Subnormal( aSig, &aExp, &aSig );
-        --aExp;
-    }
-    return packFloat64( aSign, aExp + 0x380, ( (uint64_t) aSig )<<29 );
-
-}
-
 /*----------------------------------------------------------------------------
 | Returns the result of converting the single-precision floating-point value
 | `a' to the extended double-precision floating-point format.  The conversion
@@ -3677,173 +3749,6 @@ int float32_unordered_quiet(float32 a, float32 b, float_status *status)
     return 0;
 }
 
-
-/*----------------------------------------------------------------------------
-| Returns the result of converting the double-precision floating-point value
-| `a' to the single-precision floating-point format.  The conversion is
-| performed according to the IEC/IEEE Standard for Binary Floating-Point
-| Arithmetic.
-*----------------------------------------------------------------------------*/
-
-float32 float64_to_float32(float64 a, float_status *status)
-{
-    flag aSign;
-    int aExp;
-    uint64_t aSig;
-    uint32_t zSig;
-    a = float64_squash_input_denormal(a, status);
-
-    aSig = extractFloat64Frac( a );
-    aExp = extractFloat64Exp( a );
-    aSign = extractFloat64Sign( a );
-    if ( aExp == 0x7FF ) {
-        if (aSig) {
-            return commonNaNToFloat32(float64ToCommonNaN(a, status), status);
-        }
-        return packFloat32( aSign, 0xFF, 0 );
-    }
-    shift64RightJamming( aSig, 22, &aSig );
-    zSig = aSig;
-    if ( aExp || zSig ) {
-        zSig |= 0x40000000;
-        aExp -= 0x381;
-    }
-    return roundAndPackFloat32(aSign, aExp, zSig, status);
-
-}
-
-
-/*----------------------------------------------------------------------------
-| Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
-| half-precision floating-point value, returning the result.  After being
-| shifted into the proper positions, the three fields are simply added
-| together to form the result.  This means that any integer portion of `zSig'
-| will be added into the exponent.  Since a properly normalized significand
-| will have an integer portion equal to 1, the `zExp' input should be 1 less
-| than the desired result exponent whenever `zSig' is a complete, normalized
-| significand.
-*----------------------------------------------------------------------------*/
-static float16 packFloat16(flag zSign, int zExp, uint16_t zSig)
-{
-    return make_float16(
-        (((uint32_t)zSign) << 15) + (((uint32_t)zExp) << 10) + zSig);
-}
-
-/*----------------------------------------------------------------------------
-| Takes an abstract floating-point value having sign `zSign', exponent `zExp',
-| and significand `zSig', and returns the proper half-precision floating-
-| point value corresponding to the abstract input.  Ordinarily, the abstract
-| value is simply rounded and packed into the half-precision format, with
-| the inexact exception raised if the abstract input cannot be represented
-| exactly.  However, if the abstract value is too large, the overflow and
-| inexact exceptions are raised and an infinity or maximal finite value is
-| returned.  If the abstract value is too small, the input value is rounded to
-| a subnormal number, and the underflow and inexact exceptions are raised if
-| the abstract input cannot be represented exactly as a subnormal half-
-| precision floating-point number.
-| The `ieee' flag indicates whether to use IEEE standard half precision, or
-| ARM-style "alternative representation", which omits the NaN and Inf
-| encodings in order to raise the maximum representable exponent by one.
-|     The input significand `zSig' has its binary point between bits 22
-| and 23, which is 13 bits to the left of the usual location.  This shifted
-| significand must be normalized or smaller.  If `zSig' is not normalized,
-| `zExp' must be 0; in that case, the result returned is a subnormal number,
-| and it must not require rounding.  In the usual case that `zSig' is
-| normalized, `zExp' must be 1 less than the ``true'' floating-point exponent.
-| Note the slightly odd position of the binary point in zSig compared with the
-| other roundAndPackFloat functions. This should probably be fixed if we
-| need to implement more float16 routines than just conversion.
-| The handling of underflow and overflow follows the IEC/IEEE Standard for
-| Binary Floating-Point Arithmetic.
-*----------------------------------------------------------------------------*/
-
-static float16 roundAndPackFloat16(flag zSign, int zExp,
-                                   uint32_t zSig, flag ieee,
-                                   float_status *status)
-{
-    int maxexp = ieee ? 29 : 30;
-    uint32_t mask;
-    uint32_t increment;
-    bool rounding_bumps_exp;
-    bool is_tiny = false;
-
-    /* Calculate the mask of bits of the mantissa which are not
-     * representable in half-precision and will be lost.
-     */
-    if (zExp < 1) {
-        /* Will be denormal in halfprec */
-        mask = 0x00ffffff;
-        if (zExp >= -11) {
-            mask >>= 11 + zExp;
-        }
-    } else {
-        /* Normal number in halfprec */
-        mask = 0x00001fff;
-    }
-
-    switch (status->float_rounding_mode) {
-    case float_round_nearest_even:
-        increment = (mask + 1) >> 1;
-        if ((zSig & mask) == increment) {
-            increment = zSig & (increment << 1);
-        }
-        break;
-    case float_round_ties_away:
-        increment = (mask + 1) >> 1;
-        break;
-    case float_round_up:
-        increment = zSign ? 0 : mask;
-        break;
-    case float_round_down:
-        increment = zSign ? mask : 0;
-        break;
-    default: /* round_to_zero */
-        increment = 0;
-        break;
-    }
-
-    rounding_bumps_exp = (zSig + increment >= 0x01000000);
-
-    if (zExp > maxexp || (zExp == maxexp && rounding_bumps_exp)) {
-        if (ieee) {
-            float_raise(float_flag_overflow | float_flag_inexact, status);
-            return packFloat16(zSign, 0x1f, 0);
-        } else {
-            float_raise(float_flag_invalid, status);
-            return packFloat16(zSign, 0x1f, 0x3ff);
-        }
-    }
-
-    if (zExp < 0) {
-        /* Note that flush-to-zero does not affect half-precision results */
-        is_tiny =
-            (status->float_detect_tininess == float_tininess_before_rounding)
-            || (zExp < -1)
-            || (!rounding_bumps_exp);
-    }
-    if (zSig & mask) {
-        float_raise(float_flag_inexact, status);
-        if (is_tiny) {
-            float_raise(float_flag_underflow, status);
-        }
-    }
-
-    zSig += increment;
-    if (rounding_bumps_exp) {
-        zSig >>= 1;
-        zExp++;
-    }
-
-    if (zExp < -10) {
-        return packFloat16(zSign, 0, 0);
-    }
-    if (zExp < 0) {
-        zSig >>= -zExp;
-        zExp = 0;
-    }
-    return packFloat16(zSign, zExp, zSig >> 13);
-}
-
 /*----------------------------------------------------------------------------
 | If `a' is denormal and we are in flush-to-zero mode then set the
 | input-denormal exception and return zero. Otherwise just return the value.
@@ -3859,163 +3764,6 @@ float16 float16_squash_input_denormal(float16 a, float_status *status)
     return a;
 }
 
-static void normalizeFloat16Subnormal(uint32_t aSig, int *zExpPtr,
-                                      uint32_t *zSigPtr)
-{
-    int8_t shiftCount = countLeadingZeros32(aSig) - 21;
-    *zSigPtr = aSig << shiftCount;
-    *zExpPtr = 1 - shiftCount;
-}
-
-/* Half precision floats come in two formats: standard IEEE and "ARM" format.
-   The latter gains extra exponent range by omitting the NaN/Inf encodings.  */
-
-float32 float16_to_float32(float16 a, flag ieee, float_status *status)
-{
-    flag aSign;
-    int aExp;
-    uint32_t aSig;
-
-    aSign = extractFloat16Sign(a);
-    aExp = extractFloat16Exp(a);
-    aSig = extractFloat16Frac(a);
-
-    if (aExp == 0x1f && ieee) {
-        if (aSig) {
-            return commonNaNToFloat32(float16ToCommonNaN(a, status), status);
-        }
-        return packFloat32(aSign, 0xff, 0);
-    }
-    if (aExp == 0) {
-        if (aSig == 0) {
-            return packFloat32(aSign, 0, 0);
-        }
-
-        normalizeFloat16Subnormal(aSig, &aExp, &aSig);
-        aExp--;
-    }
-    return packFloat32( aSign, aExp + 0x70, aSig << 13);
-}
-
-float16 float32_to_float16(float32 a, flag ieee, float_status *status)
-{
-    flag aSign;
-    int aExp;
-    uint32_t aSig;
-
-    a = float32_squash_input_denormal(a, status);
-
-    aSig = extractFloat32Frac( a );
-    aExp = extractFloat32Exp( a );
-    aSign = extractFloat32Sign( a );
-    if ( aExp == 0xFF ) {
-        if (aSig) {
-            /* Input is a NaN */
-            if (!ieee) {
-                float_raise(float_flag_invalid, status);
-                return packFloat16(aSign, 0, 0);
-            }
-            return commonNaNToFloat16(
-                float32ToCommonNaN(a, status), status);
-        }
-        /* Infinity */
-        if (!ieee) {
-            float_raise(float_flag_invalid, status);
-            return packFloat16(aSign, 0x1f, 0x3ff);
-        }
-        return packFloat16(aSign, 0x1f, 0);
-    }
-    if (aExp == 0 && aSig == 0) {
-        return packFloat16(aSign, 0, 0);
-    }
-    /* Decimal point between bits 22 and 23. Note that we add the 1 bit
-     * even if the input is denormal; however this is harmless because
-     * the largest possible single-precision denormal is still smaller
-     * than the smallest representable half-precision denormal, and so we
-     * will end up ignoring aSig and returning via the "always return zero"
-     * codepath.
-     */
-    aSig |= 0x00800000;
-    aExp -= 0x71;
-
-    return roundAndPackFloat16(aSign, aExp, aSig, ieee, status);
-}
-
-float64 float16_to_float64(float16 a, flag ieee, float_status *status)
-{
-    flag aSign;
-    int aExp;
-    uint32_t aSig;
-
-    aSign = extractFloat16Sign(a);
-    aExp = extractFloat16Exp(a);
-    aSig = extractFloat16Frac(a);
-
-    if (aExp == 0x1f && ieee) {
-        if (aSig) {
-            return commonNaNToFloat64(
-                float16ToCommonNaN(a, status), status);
-        }
-        return packFloat64(aSign, 0x7ff, 0);
-    }
-    if (aExp == 0) {
-        if (aSig == 0) {
-            return packFloat64(aSign, 0, 0);
-        }
-
-        normalizeFloat16Subnormal(aSig, &aExp, &aSig);
-        aExp--;
-    }
-    return packFloat64(aSign, aExp + 0x3f0, ((uint64_t)aSig) << 42);
-}
-
-float16 float64_to_float16(float64 a, flag ieee, float_status *status)
-{
-    flag aSign;
-    int aExp;
-    uint64_t aSig;
-    uint32_t zSig;
-
-    a = float64_squash_input_denormal(a, status);
-
-    aSig = extractFloat64Frac(a);
-    aExp = extractFloat64Exp(a);
-    aSign = extractFloat64Sign(a);
-    if (aExp == 0x7FF) {
-        if (aSig) {
-            /* Input is a NaN */
-            if (!ieee) {
-                float_raise(float_flag_invalid, status);
-                return packFloat16(aSign, 0, 0);
-            }
-            return commonNaNToFloat16(
-                float64ToCommonNaN(a, status), status);
-        }
-        /* Infinity */
-        if (!ieee) {
-            float_raise(float_flag_invalid, status);
-            return packFloat16(aSign, 0x1f, 0x3ff);
-        }
-        return packFloat16(aSign, 0x1f, 0);
-    }
-    shift64RightJamming(aSig, 29, &aSig);
-    zSig = aSig;
-    if (aExp == 0 && zSig == 0) {
-        return packFloat16(aSign, 0, 0);
-    }
-    /* Decimal point between bits 22 and 23. Note that we add the 1 bit
-     * even if the input is denormal; however this is harmless because
-     * the largest possible single-precision denormal is still smaller
-     * than the smallest representable half-precision denormal, and so we
-     * will end up ignoring aSig and returning via the "always return zero"
-     * codepath.
-     */
-    zSig |= 0x00800000;
-    aExp -= 0x3F1;
-
-    return roundAndPackFloat16(aSign, aExp, zSig, ieee, status);
-}
-
 /*----------------------------------------------------------------------------
 | Returns the result of converting the double-precision floating-point value
 | `a' to the extended double-precision floating-point format.  The conversion
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 14/28] fpu/softfloat: re-factor float to float conversions

[Peter Maydell]

On 14 May 2018 at 23:12, Richard Henderson <richard.henderson@linaro.org> wrote:
> From: Alex Bennée <alex.bennee@linaro.org>
>
> This allows us to delete a lot of additional boilerplate
> code which is no longer needed.
>
> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>

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

thanks
-- PMM

[Qemu-devel] [PATCH v5 15/28] target/hppa: Remove floatX_maybe_silence_nan from conversions

[Richard Henderson]

This is now handled properly by the generic softfloat code.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/hppa/op_helper.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/target/hppa/op_helper.c b/target/hppa/op_helper.c
index a3af62daf7..912e8d5be4 100644
--- a/target/hppa/op_helper.c
+++ b/target/hppa/op_helper.c
@@ -341,7 +341,6 @@ float64 HELPER(fdiv_d)(CPUHPPAState *env, float64 a, float64 b)
 float64 HELPER(fcnv_s_d)(CPUHPPAState *env, float32 arg)
 {
     float64 ret = float32_to_float64(arg, &env->fp_status);
-    ret = float64_maybe_silence_nan(ret, &env->fp_status);
     update_fr0_op(env, GETPC());
     return ret;
 }
@@ -349,7 +348,6 @@ float64 HELPER(fcnv_s_d)(CPUHPPAState *env, float32 arg)
 float32 HELPER(fcnv_d_s)(CPUHPPAState *env, float64 arg)
 {
     float32 ret = float64_to_float32(arg, &env->fp_status);
-    ret = float32_maybe_silence_nan(ret, &env->fp_status);
     update_fr0_op(env, GETPC());
     return ret;
 }
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 15/28] target/hppa: Remove floatX_maybe_silence_nan from conversions

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> This is now handled properly by the generic softfloat code.
>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  target/hppa/op_helper.c | 2 --
>  1 file changed, 2 deletions(-)
>
> diff --git a/target/hppa/op_helper.c b/target/hppa/op_helper.c
> index a3af62daf7..912e8d5be4 100644
> --- a/target/hppa/op_helper.c
> +++ b/target/hppa/op_helper.c
> @@ -341,7 +341,6 @@ float64 HELPER(fdiv_d)(CPUHPPAState *env, float64 a, float64 b)
>  float64 HELPER(fcnv_s_d)(CPUHPPAState *env, float32 arg)
>  {
>      float64 ret = float32_to_float64(arg, &env->fp_status);
> -    ret = float64_maybe_silence_nan(ret, &env->fp_status);
>      update_fr0_op(env, GETPC());
>      return ret;
>  }
> @@ -349,7 +348,6 @@ float64 HELPER(fcnv_s_d)(CPUHPPAState *env, float32 arg)
>  float32 HELPER(fcnv_d_s)(CPUHPPAState *env, float64 arg)
>  {
>      float32 ret = float64_to_float32(arg, &env->fp_status);
> -    ret = float32_maybe_silence_nan(ret, &env->fp_status);
>      update_fr0_op(env, GETPC());
>      return ret;
>  }


--
Alex Bennée

[Qemu-devel] [PATCH v5 16/28] target/m68k: Use floatX_silence_nan when we have already checked for SNaN

[Richard Henderson]

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/m68k/softfloat.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/target/m68k/softfloat.c b/target/m68k/softfloat.c
index e41b07d042..6ec227e20f 100644
--- a/target/m68k/softfloat.c
+++ b/target/m68k/softfloat.c
@@ -31,13 +31,14 @@ static floatx80 propagateFloatx80NaNOneArg(floatx80 a, float_status *status)
 {
     if (floatx80_is_signaling_nan(a, status)) {
         float_raise(float_flag_invalid, status);
+        a = floatx80_silence_nan(a, status);
     }
 
     if (status->default_nan_mode) {
         return floatx80_default_nan(status);
     }
 
-    return floatx80_maybe_silence_nan(a, status);
+    return a;
 }
 
 /*----------------------------------------------------------------------------
-- 
2.17.0

[Qemu-devel] [PATCH v5 17/28] target/mips: Remove floatX_maybe_silence_nan from conversions

[Richard Henderson]

This is now handled properly by the generic softfloat code.

Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Yongbok Kim <yongbok.kim@mips.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/mips/msa_helper.c | 4 ----
 target/mips/op_helper.c  | 2 --
 2 files changed, 6 deletions(-)

diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index 8fb7a369ca..c74e3cdc65 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -1615,7 +1615,6 @@ static inline float16 float16_from_float32(int32_t a, flag ieee,
       float16 f_val;
 
       f_val = float32_to_float16((float32)a, ieee, status);
-      f_val = float16_maybe_silence_nan(f_val, status);
 
       return a < 0 ? (f_val | (1 << 15)) : f_val;
 }
@@ -1625,7 +1624,6 @@ static inline float32 float32_from_float64(int64_t a, float_status *status)
       float32 f_val;
 
       f_val = float64_to_float32((float64)a, status);
-      f_val = float32_maybe_silence_nan(f_val, status);
 
       return a < 0 ? (f_val | (1 << 31)) : f_val;
 }
@@ -1636,7 +1634,6 @@ static inline float32 float32_from_float16(int16_t a, flag ieee,
       float32 f_val;
 
       f_val = float16_to_float32((float16)a, ieee, status);
-      f_val = float32_maybe_silence_nan(f_val, status);
 
       return a < 0 ? (f_val | (1 << 31)) : f_val;
 }
@@ -1646,7 +1643,6 @@ static inline float64 float64_from_float32(int32_t a, float_status *status)
       float64 f_val;
 
       f_val = float32_to_float64((float64)a, status);
-      f_val = float64_maybe_silence_nan(f_val, status);
 
       return a < 0 ? (f_val | (1ULL << 63)) : f_val;
 }
diff --git a/target/mips/op_helper.c b/target/mips/op_helper.c
index 798cdad030..9025f42366 100644
--- a/target/mips/op_helper.c
+++ b/target/mips/op_helper.c
@@ -2700,7 +2700,6 @@ uint64_t helper_float_cvtd_s(CPUMIPSState *env, uint32_t fst0)
     uint64_t fdt2;
 
     fdt2 = float32_to_float64(fst0, &env->active_fpu.fp_status);
-    fdt2 = float64_maybe_silence_nan(fdt2, &env->active_fpu.fp_status);
     update_fcr31(env, GETPC());
     return fdt2;
 }
@@ -2790,7 +2789,6 @@ uint32_t helper_float_cvts_d(CPUMIPSState *env, uint64_t fdt0)
     uint32_t fst2;
 
     fst2 = float64_to_float32(fdt0, &env->active_fpu.fp_status);
-    fst2 = float32_maybe_silence_nan(fst2, &env->active_fpu.fp_status);
     update_fcr31(env, GETPC());
     return fst2;
 }
-- 
2.17.0

[Qemu-devel] [PATCH v5 18/28] target/riscv: Remove floatX_maybe_silence_nan from conversions

[Richard Henderson]

This is now handled properly by the generic softfloat code.

Cc: Palmer Dabbelt <palmer@sifive.com>
Cc: Sagar Karandikar <sagark@eecs.berkeley.edu>
Cc: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Michael Clark <mjc@sifive.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/riscv/fpu_helper.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/target/riscv/fpu_helper.c b/target/riscv/fpu_helper.c
index abbadead5c..fdb87d8d82 100644
--- a/target/riscv/fpu_helper.c
+++ b/target/riscv/fpu_helper.c
@@ -279,14 +279,12 @@ uint64_t helper_fmax_d(CPURISCVState *env, uint64_t frs1, uint64_t frs2)
 
 uint64_t helper_fcvt_s_d(CPURISCVState *env, uint64_t rs1)
 {
-    rs1 = float64_to_float32(rs1, &env->fp_status);
-    return float32_maybe_silence_nan(rs1, &env->fp_status);
+    return float64_to_float32(rs1, &env->fp_status);
 }
 
 uint64_t helper_fcvt_d_s(CPURISCVState *env, uint64_t rs1)
 {
-    rs1 = float32_to_float64(rs1, &env->fp_status);
-    return float64_maybe_silence_nan(rs1, &env->fp_status);
+    return float32_to_float64(rs1, &env->fp_status);
 }
 
 uint64_t helper_fsqrt_d(CPURISCVState *env, uint64_t frs1)
-- 
2.17.0

[Qemu-devel] [PATCH v5 19/28] target/s390x: Remove floatX_maybe_silence_nan from conversions

[Richard Henderson]

This is now handled properly by the generic softfloat code.

Cc: Alexander Graf <agraf@suse.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/s390x/fpu_helper.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/target/s390x/fpu_helper.c b/target/s390x/fpu_helper.c
index 43f8bf1c94..5c5b451b3b 100644
--- a/target/s390x/fpu_helper.c
+++ b/target/s390x/fpu_helper.c
@@ -269,7 +269,7 @@ uint64_t HELPER(ldeb)(CPUS390XState *env, uint64_t f2)
 {
     float64 ret = float32_to_float64(f2, &env->fpu_status);
     handle_exceptions(env, GETPC());
-    return float64_maybe_silence_nan(ret, &env->fpu_status);
+    return ret;
 }
 
 /* convert 128-bit float to 64-bit float */
@@ -277,7 +277,7 @@ uint64_t HELPER(ldxb)(CPUS390XState *env, uint64_t ah, uint64_t al)
 {
     float64 ret = float128_to_float64(make_float128(ah, al), &env->fpu_status);
     handle_exceptions(env, GETPC());
-    return float64_maybe_silence_nan(ret, &env->fpu_status);
+    return ret;
 }
 
 /* convert 64-bit float to 128-bit float */
@@ -285,7 +285,7 @@ uint64_t HELPER(lxdb)(CPUS390XState *env, uint64_t f2)
 {
     float128 ret = float64_to_float128(f2, &env->fpu_status);
     handle_exceptions(env, GETPC());
-    return RET128(float128_maybe_silence_nan(ret, &env->fpu_status));
+    return RET128(ret);
 }
 
 /* convert 32-bit float to 128-bit float */
@@ -293,7 +293,7 @@ uint64_t HELPER(lxeb)(CPUS390XState *env, uint64_t f2)
 {
     float128 ret = float32_to_float128(f2, &env->fpu_status);
     handle_exceptions(env, GETPC());
-    return RET128(float128_maybe_silence_nan(ret, &env->fpu_status));
+    return RET128(ret);
 }
 
 /* convert 64-bit float to 32-bit float */
@@ -301,7 +301,7 @@ uint64_t HELPER(ledb)(CPUS390XState *env, uint64_t f2)
 {
     float32 ret = float64_to_float32(f2, &env->fpu_status);
     handle_exceptions(env, GETPC());
-    return float32_maybe_silence_nan(ret, &env->fpu_status);
+    return ret;
 }
 
 /* convert 128-bit float to 32-bit float */
@@ -309,7 +309,7 @@ uint64_t HELPER(lexb)(CPUS390XState *env, uint64_t ah, uint64_t al)
 {
     float32 ret = float128_to_float32(make_float128(ah, al), &env->fpu_status);
     handle_exceptions(env, GETPC());
-    return float32_maybe_silence_nan(ret, &env->fpu_status);
+    return ret;
 }
 
 /* 32-bit FP compare */
-- 
2.17.0

[Qemu-devel] [PATCH v5 20/28] fpu/softfloat: Use float*_silence_nan in propagateFloat*NaN

[Richard Henderson]

We have already checked the arguments for SNaN;
we don't need to do it again.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 44 +++++++++++++++++++++++++++++---------
 1 file changed, 34 insertions(+), 10 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 995a0132c6..4fa068a5dc 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -498,7 +498,7 @@ static float32 commonNaNToFloat32(commonNaNT a, float_status *status)
 | The routine is passed various bits of information about the
 | two NaNs and should return 0 to select NaN a and 1 for NaN b.
 | Note that signalling NaNs are always squashed to quiet NaNs
-| by the caller, by calling floatXX_maybe_silence_nan() before
+| by the caller, by calling floatXX_silence_nan() before
 | returning them.
 |
 | aIsLargerSignificand is only valid if both a and b are NaNs
@@ -536,7 +536,7 @@ static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
 {
     /* According to MIPS specifications, if one of the two operands is
      * a sNaN, a new qNaN has to be generated. This is done in
-     * floatXX_maybe_silence_nan(). For qNaN inputs the specifications
+     * floatXX_silence_nan(). For qNaN inputs the specifications
      * says: "When possible, this QNaN result is one of the operand QNaN
      * values." In practice it seems that most implementations choose
      * the first operand if both operands are qNaN. In short this gives
@@ -788,9 +788,15 @@ static float32 propagateFloat32NaN(float32 a, float32 b, float_status *status)
 
     if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
                 aIsLargerSignificand)) {
-        return float32_maybe_silence_nan(b, status);
+        if (bIsSignalingNaN) {
+            return float32_silence_nan(b, status);
+        }
+        return b;
     } else {
-        return float32_maybe_silence_nan(a, status);
+        if (aIsSignalingNaN) {
+            return float32_silence_nan(a, status);
+        }
+        return a;
     }
 }
 
@@ -950,9 +956,15 @@ static float64 propagateFloat64NaN(float64 a, float64 b, float_status *status)
 
     if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
                 aIsLargerSignificand)) {
-        return float64_maybe_silence_nan(b, status);
+        if (bIsSignalingNaN) {
+            return float64_silence_nan(b, status);
+        }
+        return b;
     } else {
-        return float64_maybe_silence_nan(a, status);
+        if (aIsSignalingNaN) {
+            return float64_silence_nan(a, status);
+        }
+        return a;
     }
 }
 
@@ -1121,9 +1133,15 @@ floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status)
 
     if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
                 aIsLargerSignificand)) {
-        return floatx80_maybe_silence_nan(b, status);
+        if (bIsSignalingNaN) {
+            return floatx80_silence_nan(b, status);
+        }
+        return b;
     } else {
-        return floatx80_maybe_silence_nan(a, status);
+        if (aIsSignalingNaN) {
+            return floatx80_silence_nan(a, status);
+        }
+        return a;
     }
 }
 
@@ -1270,8 +1288,14 @@ static float128 propagateFloat128NaN(float128 a, float128 b,
 
     if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
                 aIsLargerSignificand)) {
-        return float128_maybe_silence_nan(b, status);
+        if (bIsSignalingNaN) {
+            return float128_silence_nan(b, status);
+        }
+        return b;
     } else {
-        return float128_maybe_silence_nan(a, status);
+        if (aIsSignalingNaN) {
+            return float128_silence_nan(a, status);
+        }
+        return a;
     }
 }
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 20/28] fpu/softfloat: Use float*_silence_nan in propagateFloat*NaN

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> We have already checked the arguments for SNaN;
> we don't need to do it again.
>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat-specialize.h | 44 +++++++++++++++++++++++++++++---------
>  1 file changed, 34 insertions(+), 10 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index 995a0132c6..4fa068a5dc 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -498,7 +498,7 @@ static float32 commonNaNToFloat32(commonNaNT a, float_status *status)
>  | The routine is passed various bits of information about the
>  | two NaNs and should return 0 to select NaN a and 1 for NaN b.
>  | Note that signalling NaNs are always squashed to quiet NaNs
> -| by the caller, by calling floatXX_maybe_silence_nan() before
> +| by the caller, by calling floatXX_silence_nan() before
>  | returning them.
>  |
>  | aIsLargerSignificand is only valid if both a and b are NaNs
> @@ -536,7 +536,7 @@ static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
>  {
>      /* According to MIPS specifications, if one of the two operands is
>       * a sNaN, a new qNaN has to be generated. This is done in
> -     * floatXX_maybe_silence_nan(). For qNaN inputs the specifications
> +     * floatXX_silence_nan(). For qNaN inputs the specifications
>       * says: "When possible, this QNaN result is one of the operand QNaN
>       * values." In practice it seems that most implementations choose
>       * the first operand if both operands are qNaN. In short this gives
> @@ -788,9 +788,15 @@ static float32 propagateFloat32NaN(float32 a, float32 b, float_status *status)
>
>      if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
>                  aIsLargerSignificand)) {
> -        return float32_maybe_silence_nan(b, status);
> +        if (bIsSignalingNaN) {
> +            return float32_silence_nan(b, status);
> +        }
> +        return b;
>      } else {
> -        return float32_maybe_silence_nan(a, status);
> +        if (aIsSignalingNaN) {
> +            return float32_silence_nan(a, status);
> +        }
> +        return a;
>      }
>  }
>
> @@ -950,9 +956,15 @@ static float64 propagateFloat64NaN(float64 a, float64 b, float_status *status)
>
>      if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
>                  aIsLargerSignificand)) {
> -        return float64_maybe_silence_nan(b, status);
> +        if (bIsSignalingNaN) {
> +            return float64_silence_nan(b, status);
> +        }
> +        return b;
>      } else {
> -        return float64_maybe_silence_nan(a, status);
> +        if (aIsSignalingNaN) {
> +            return float64_silence_nan(a, status);
> +        }
> +        return a;
>      }
>  }
>
> @@ -1121,9 +1133,15 @@ floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status)
>
>      if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
>                  aIsLargerSignificand)) {
> -        return floatx80_maybe_silence_nan(b, status);
> +        if (bIsSignalingNaN) {
> +            return floatx80_silence_nan(b, status);
> +        }
> +        return b;
>      } else {
> -        return floatx80_maybe_silence_nan(a, status);
> +        if (aIsSignalingNaN) {
> +            return floatx80_silence_nan(a, status);
> +        }
> +        return a;
>      }
>  }
>
> @@ -1270,8 +1288,14 @@ static float128 propagateFloat128NaN(float128 a, float128 b,
>
>      if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
>                  aIsLargerSignificand)) {
> -        return float128_maybe_silence_nan(b, status);
> +        if (bIsSignalingNaN) {
> +            return float128_silence_nan(b, status);
> +        }
> +        return b;
>      } else {
> -        return float128_maybe_silence_nan(a, status);
> +        if (aIsSignalingNaN) {
> +            return float128_silence_nan(a, status);
> +        }
> +        return a;
>      }
>  }


--
Alex Bennée

[Qemu-devel] [PATCH v5 21/28] fpu/softfloat: Remove floatX_maybe_silence_nan

[Richard Henderson]

These functions are now unused.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 63 --------------------------------------
 include/fpu/softfloat.h    |  5 ---
 2 files changed, 68 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 4fa068a5dc..d7033b7757 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -364,19 +364,6 @@ float16 float16_silence_nan(float16 a, float_status *status)
 #endif
 }
 
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN if the half-precision floating point value `a' is a
-| signaling NaN; otherwise returns `a'.
-*----------------------------------------------------------------------------*/
-
-float16 float16_maybe_silence_nan(float16 a, float_status *status)
-{
-    if (float16_is_signaling_nan(a, status)) {
-        return float16_silence_nan(a, status);
-    }
-    return a;
-}
-
 /*----------------------------------------------------------------------------
 | Returns 1 if the single-precision floating-point value `a' is a quiet
 | NaN; otherwise returns 0.
@@ -438,18 +425,6 @@ float32 float32_silence_nan(float32 a, float_status *status)
     }
 #endif
 }
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN if the single-precision floating point value `a' is a
-| signaling NaN; otherwise returns `a'.
-*----------------------------------------------------------------------------*/
-
-float32 float32_maybe_silence_nan(float32 a, float_status *status)
-{
-    if (float32_is_signaling_nan(a, status)) {
-        return float32_silence_nan(a, status);
-    }
-    return a;
-}
 
 /*----------------------------------------------------------------------------
 | Returns the result of converting the single-precision floating-point NaN
@@ -864,18 +839,6 @@ float64 float64_silence_nan(float64 a, float_status *status)
 #endif
 }
 
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN if the double-precision floating point value `a' is a
-| signaling NaN; otherwise returns `a'.
-*----------------------------------------------------------------------------*/
-
-float64 float64_maybe_silence_nan(float64 a, float_status *status)
-{
-    if (float64_is_signaling_nan(a, status)) {
-        return float64_silence_nan(a, status);
-    }
-    return a;
-}
 
 /*----------------------------------------------------------------------------
 | Returns the result of converting the double-precision floating-point NaN
@@ -1037,19 +1000,6 @@ floatx80 floatx80_silence_nan(floatx80 a, float_status *status)
 #endif
 }
 
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN if the extended double-precision floating point value
-| `a' is a signaling NaN; otherwise returns `a'.
-*----------------------------------------------------------------------------*/
-
-floatx80 floatx80_maybe_silence_nan(floatx80 a, float_status *status)
-{
-    if (floatx80_is_signaling_nan(a, status)) {
-        return floatx80_silence_nan(a, status);
-    }
-    return a;
-}
-
 /*----------------------------------------------------------------------------
 | Returns the result of converting the extended double-precision floating-
 | point NaN `a' to the canonical NaN format.  If `a' is a signaling NaN, the
@@ -1204,19 +1154,6 @@ float128 float128_silence_nan(float128 a, float_status *status)
 #endif
 }
 
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN if the quadruple-precision floating point value `a' is
-| a signaling NaN; otherwise returns `a'.
-*----------------------------------------------------------------------------*/
-
-float128 float128_maybe_silence_nan(float128 a, float_status *status)
-{
-    if (float128_is_signaling_nan(a, status)) {
-        return float128_silence_nan(a, status);
-    }
-    return a;
-}
-
 /*----------------------------------------------------------------------------
 | Returns the result of converting the quadruple-precision floating-point NaN
 | `a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h
index a6860e858d..69f4dbc4db 100644
--- a/include/fpu/softfloat.h
+++ b/include/fpu/softfloat.h
@@ -258,7 +258,6 @@ int float16_compare_quiet(float16, float16, float_status *status);
 int float16_is_quiet_nan(float16, float_status *status);
 int float16_is_signaling_nan(float16, float_status *status);
 float16 float16_silence_nan(float16, float_status *status);
-float16 float16_maybe_silence_nan(float16, float_status *status);
 
 static inline int float16_is_any_nan(float16 a)
 {
@@ -370,7 +369,6 @@ float32 float32_maxnummag(float32, float32, float_status *status);
 int float32_is_quiet_nan(float32, float_status *status);
 int float32_is_signaling_nan(float32, float_status *status);
 float32 float32_silence_nan(float32, float_status *status);
-float32 float32_maybe_silence_nan(float32, float_status *status);
 float32 float32_scalbn(float32, int, float_status *status);
 
 static inline float32 float32_abs(float32 a)
@@ -500,7 +498,6 @@ float64 float64_maxnummag(float64, float64, float_status *status);
 int float64_is_quiet_nan(float64 a, float_status *status);
 int float64_is_signaling_nan(float64, float_status *status);
 float64 float64_silence_nan(float64, float_status *status);
-float64 float64_maybe_silence_nan(float64, float_status *status);
 float64 float64_scalbn(float64, int, float_status *status);
 
 static inline float64 float64_abs(float64 a)
@@ -604,7 +601,6 @@ int floatx80_compare_quiet(floatx80, floatx80, float_status *status);
 int floatx80_is_quiet_nan(floatx80, float_status *status);
 int floatx80_is_signaling_nan(floatx80, float_status *status);
 floatx80 floatx80_silence_nan(floatx80, float_status *status);
-floatx80 floatx80_maybe_silence_nan(floatx80, float_status *status);
 floatx80 floatx80_scalbn(floatx80, int, float_status *status);
 
 static inline floatx80 floatx80_abs(floatx80 a)
@@ -816,7 +812,6 @@ int float128_compare_quiet(float128, float128, float_status *status);
 int float128_is_quiet_nan(float128, float_status *status);
 int float128_is_signaling_nan(float128, float_status *status);
 float128 float128_silence_nan(float128, float_status *status);
-float128 float128_maybe_silence_nan(float128, float_status *status);
 float128 float128_scalbn(float128, int, float_status *status);
 
 static inline float128 float128_abs(float128 a)
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 21/28] fpu/softfloat: Remove floatX_maybe_silence_nan

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> These functions are now unused.
>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat-specialize.h | 63 --------------------------------------
>  include/fpu/softfloat.h    |  5 ---
>  2 files changed, 68 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index 4fa068a5dc..d7033b7757 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -364,19 +364,6 @@ float16 float16_silence_nan(float16 a, float_status *status)
>  #endif
>  }
>
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN if the half-precision floating point value `a' is a
> -| signaling NaN; otherwise returns `a'.
> -*----------------------------------------------------------------------------*/
> -
> -float16 float16_maybe_silence_nan(float16 a, float_status *status)
> -{
> -    if (float16_is_signaling_nan(a, status)) {
> -        return float16_silence_nan(a, status);
> -    }
> -    return a;
> -}
> -
>  /*----------------------------------------------------------------------------
>  | Returns 1 if the single-precision floating-point value `a' is a quiet
>  | NaN; otherwise returns 0.
> @@ -438,18 +425,6 @@ float32 float32_silence_nan(float32 a, float_status *status)
>      }
>  #endif
>  }
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN if the single-precision floating point value `a' is a
> -| signaling NaN; otherwise returns `a'.
> -*----------------------------------------------------------------------------*/
> -
> -float32 float32_maybe_silence_nan(float32 a, float_status *status)
> -{
> -    if (float32_is_signaling_nan(a, status)) {
> -        return float32_silence_nan(a, status);
> -    }
> -    return a;
> -}
>
>  /*----------------------------------------------------------------------------
>  | Returns the result of converting the single-precision floating-point NaN
> @@ -864,18 +839,6 @@ float64 float64_silence_nan(float64 a, float_status *status)
>  #endif
>  }
>
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN if the double-precision floating point value `a' is a
> -| signaling NaN; otherwise returns `a'.
> -*----------------------------------------------------------------------------*/
> -
> -float64 float64_maybe_silence_nan(float64 a, float_status *status)
> -{
> -    if (float64_is_signaling_nan(a, status)) {
> -        return float64_silence_nan(a, status);
> -    }
> -    return a;
> -}
>
>  /*----------------------------------------------------------------------------
>  | Returns the result of converting the double-precision floating-point NaN
> @@ -1037,19 +1000,6 @@ floatx80 floatx80_silence_nan(floatx80 a, float_status *status)
>  #endif
>  }
>
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN if the extended double-precision floating point value
> -| `a' is a signaling NaN; otherwise returns `a'.
> -*----------------------------------------------------------------------------*/
> -
> -floatx80 floatx80_maybe_silence_nan(floatx80 a, float_status *status)
> -{
> -    if (floatx80_is_signaling_nan(a, status)) {
> -        return floatx80_silence_nan(a, status);
> -    }
> -    return a;
> -}
> -
>  /*----------------------------------------------------------------------------
>  | Returns the result of converting the extended double-precision floating-
>  | point NaN `a' to the canonical NaN format.  If `a' is a signaling NaN, the
> @@ -1204,19 +1154,6 @@ float128 float128_silence_nan(float128 a, float_status *status)
>  #endif
>  }
>
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN if the quadruple-precision floating point value `a' is
> -| a signaling NaN; otherwise returns `a'.
> -*----------------------------------------------------------------------------*/
> -
> -float128 float128_maybe_silence_nan(float128 a, float_status *status)
> -{
> -    if (float128_is_signaling_nan(a, status)) {
> -        return float128_silence_nan(a, status);
> -    }
> -    return a;
> -}
> -
>  /*----------------------------------------------------------------------------
>  | Returns the result of converting the quadruple-precision floating-point NaN
>  | `a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
> diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h
> index a6860e858d..69f4dbc4db 100644
> --- a/include/fpu/softfloat.h
> +++ b/include/fpu/softfloat.h
> @@ -258,7 +258,6 @@ int float16_compare_quiet(float16, float16, float_status *status);
>  int float16_is_quiet_nan(float16, float_status *status);
>  int float16_is_signaling_nan(float16, float_status *status);
>  float16 float16_silence_nan(float16, float_status *status);
> -float16 float16_maybe_silence_nan(float16, float_status *status);
>
>  static inline int float16_is_any_nan(float16 a)
>  {
> @@ -370,7 +369,6 @@ float32 float32_maxnummag(float32, float32, float_status *status);
>  int float32_is_quiet_nan(float32, float_status *status);
>  int float32_is_signaling_nan(float32, float_status *status);
>  float32 float32_silence_nan(float32, float_status *status);
> -float32 float32_maybe_silence_nan(float32, float_status *status);
>  float32 float32_scalbn(float32, int, float_status *status);
>
>  static inline float32 float32_abs(float32 a)
> @@ -500,7 +498,6 @@ float64 float64_maxnummag(float64, float64, float_status *status);
>  int float64_is_quiet_nan(float64 a, float_status *status);
>  int float64_is_signaling_nan(float64, float_status *status);
>  float64 float64_silence_nan(float64, float_status *status);
> -float64 float64_maybe_silence_nan(float64, float_status *status);
>  float64 float64_scalbn(float64, int, float_status *status);
>
>  static inline float64 float64_abs(float64 a)
> @@ -604,7 +601,6 @@ int floatx80_compare_quiet(floatx80, floatx80, float_status *status);
>  int floatx80_is_quiet_nan(floatx80, float_status *status);
>  int floatx80_is_signaling_nan(floatx80, float_status *status);
>  floatx80 floatx80_silence_nan(floatx80, float_status *status);
> -floatx80 floatx80_maybe_silence_nan(floatx80, float_status *status);
>  floatx80 floatx80_scalbn(floatx80, int, float_status *status);
>
>  static inline floatx80 floatx80_abs(floatx80 a)
> @@ -816,7 +812,6 @@ int float128_compare_quiet(float128, float128, float_status *status);
>  int float128_is_quiet_nan(float128, float_status *status);
>  int float128_is_signaling_nan(float128, float_status *status);
>  float128 float128_silence_nan(float128, float_status *status);
> -float128 float128_maybe_silence_nan(float128, float_status *status);
>  float128 float128_scalbn(float128, int, float_status *status);
>
>  static inline float128 float128_abs(float128 a)


--
Alex Bennée

[Qemu-devel] [PATCH v5 22/28] fpu/softfloat: Specialize on snan_bit_is_one

[Richard Henderson]

Only MIPS requires snan_bit_is_one to be variable.  While we are
specializing softfloat behaviour, allow other targets to eliminate
this runtime check.

Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Yongbok Kim <yongbok.kim@mips.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Alexander Graf <agraf@suse.de>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
v5
  - do not remove the set_snan_bit_is_one function
  - tidy the language in the snan_bit_is_one block comment
---
 fpu/softfloat-specialize.h    | 68 ++++++++++++++++++++++-------------
 include/fpu/softfloat-types.h |  1 +
 target/hppa/cpu.c             |  1 -
 target/ppc/fpu_helper.c       |  1 -
 target/sh4/cpu.c              |  1 -
 target/unicore32/cpu.c        |  2 --
 6 files changed, 44 insertions(+), 30 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index d7033b7757..d1e06da75b 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -79,13 +79,31 @@ this code that are retained.
  * version 2 or later. See the COPYING file in the top-level directory.
  */
 
-#if defined(TARGET_XTENSA)
 /* Define for architectures which deviate from IEEE in not supporting
  * signaling NaNs (so all NaNs are treated as quiet).
  */
+#if defined(TARGET_XTENSA)
 #define NO_SIGNALING_NANS 1
 #endif
 
+/* Define how the architecture discriminates signaling NaNs.
+ * This done with the most significant bit of the fraction.
+ * In IEEE 754-1985 this was implementation defined, but in IEEE 754-2008
+ * the msb must be zero.  MIPS is (so far) unique in supporting both the
+ * 2008 revision and backward compatibility with their original choice.
+ * Thus for MIPS we must make the choice at runtime.
+ */
+static inline flag snan_bit_is_one(float_status *status)
+{
+#if defined(TARGET_MIPS)
+    return status->snan_bit_is_one;
+#elif defined(TARGET_HPPA) || defined(TARGET_UNICORE32) || defined(TARGET_SH4)
+    return 1;
+#else
+    return 0;
+#endif
+}
+
 /*----------------------------------------------------------------------------
 | For the deconstructed floating-point with fraction FRAC, return true
 | if the fraction represents a signalling NaN; otherwise false.
@@ -97,7 +115,7 @@ static bool parts_is_snan_frac(uint64_t frac, float_status *status)
     return false;
 #else
     flag msb = extract64(frac, DECOMPOSED_BINARY_POINT - 1, 1);
-    return msb == status->snan_bit_is_one;
+    return msb == snan_bit_is_one(status);
 #endif
 }
 
@@ -118,7 +136,7 @@ static FloatParts parts_default_nan(float_status *status)
 #elif defined(TARGET_HPPA)
     frac = 1ULL << (DECOMPOSED_BINARY_POINT - 2);
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         frac = (1ULL << (DECOMPOSED_BINARY_POINT - 1)) - 1;
     } else {
 #if defined(TARGET_MIPS)
@@ -151,7 +169,7 @@ static FloatParts parts_silence_nan(FloatParts a, float_status *status)
     a.frac &= ~(1ULL << (DECOMPOSED_BINARY_POINT - 1));
     a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 2);
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return parts_default_nan(status);
     } else {
         a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 1);
@@ -169,7 +187,7 @@ float16 float16_default_nan(float_status *status)
 #if defined(TARGET_ARM)
     return const_float16(0x7E00);
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return const_float16(0x7DFF);
     } else {
 #if defined(TARGET_MIPS)
@@ -195,7 +213,7 @@ float32 float32_default_nan(float_status *status)
 #elif defined(TARGET_HPPA)
     return const_float32(0x7FA00000);
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return const_float32(0x7FBFFFFF);
     } else {
 #if defined(TARGET_MIPS)
@@ -220,7 +238,7 @@ float64 float64_default_nan(float_status *status)
 #elif defined(TARGET_HPPA)
     return const_float64(LIT64(0x7FF4000000000000));
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return const_float64(LIT64(0x7FF7FFFFFFFFFFFF));
     } else {
 #if defined(TARGET_MIPS)
@@ -242,7 +260,7 @@ floatx80 floatx80_default_nan(float_status *status)
     r.low = LIT64(0xFFFFFFFFFFFFFFFF);
     r.high = 0x7FFF;
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         r.low = LIT64(0xBFFFFFFFFFFFFFFF);
         r.high = 0x7FFF;
     } else {
@@ -274,7 +292,7 @@ float128 float128_default_nan(float_status *status)
 {
     float128 r;
 
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         r.low = LIT64(0xFFFFFFFFFFFFFFFF);
         r.high = LIT64(0x7FFF7FFFFFFFFFFF);
     } else {
@@ -319,7 +337,7 @@ int float16_is_quiet_nan(float16 a_, float_status *status)
     return float16_is_any_nan(a_);
 #else
     uint16_t a = float16_val(a_);
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return (((a >> 9) & 0x3F) == 0x3E) && (a & 0x1FF);
     } else {
         return ((a & ~0x8000) >= 0x7C80);
@@ -338,7 +356,7 @@ int float16_is_signaling_nan(float16 a_, float_status *status)
     return 0;
 #else
     uint16_t a = float16_val(a_);
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return ((a & ~0x8000) >= 0x7C80);
     } else {
         return (((a >> 9) & 0x3F) == 0x3E) && (a & 0x1FF);
@@ -356,7 +374,7 @@ float16 float16_silence_nan(float16 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     g_assert_not_reached();
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return float16_default_nan(status);
     } else {
         return a | (1 << 9);
@@ -375,7 +393,7 @@ int float32_is_quiet_nan(float32 a_, float_status *status)
     return float32_is_any_nan(a_);
 #else
     uint32_t a = float32_val(a_);
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return (((a >> 22) & 0x1FF) == 0x1FE) && (a & 0x003FFFFF);
     } else {
         return ((uint32_t)(a << 1) >= 0xFF800000);
@@ -394,7 +412,7 @@ int float32_is_signaling_nan(float32 a_, float_status *status)
     return 0;
 #else
     uint32_t a = float32_val(a_);
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return ((uint32_t)(a << 1) >= 0xFF800000);
     } else {
         return (((a >> 22) & 0x1FF) == 0x1FE) && (a & 0x003FFFFF);
@@ -412,7 +430,7 @@ float32 float32_silence_nan(float32 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     g_assert_not_reached();
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
 # ifdef TARGET_HPPA
         a &= ~0x00400000;
         a |=  0x00200000;
@@ -651,7 +669,7 @@ static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
         return 3;
     }
 
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         /* Prefer sNaN over qNaN, in the a, b, c order. */
         if (aIsSNaN) {
             return 0;
@@ -786,7 +804,7 @@ int float64_is_quiet_nan(float64 a_, float_status *status)
     return float64_is_any_nan(a_);
 #else
     uint64_t a = float64_val(a_);
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return (((a >> 51) & 0xFFF) == 0xFFE)
             && (a & 0x0007FFFFFFFFFFFFULL);
     } else {
@@ -806,7 +824,7 @@ int float64_is_signaling_nan(float64 a_, float_status *status)
     return 0;
 #else
     uint64_t a = float64_val(a_);
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return ((a << 1) >= 0xFFF0000000000000ULL);
     } else {
         return (((a >> 51) & 0xFFF) == 0xFFE)
@@ -825,7 +843,7 @@ float64 float64_silence_nan(float64 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     g_assert_not_reached();
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
 # ifdef TARGET_HPPA
         a &= ~0x0008000000000000ULL;
         a |=  0x0004000000000000ULL;
@@ -942,7 +960,7 @@ int floatx80_is_quiet_nan(floatx80 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     return floatx80_is_any_nan(a);
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         uint64_t aLow;
 
         aLow = a.low & ~0x4000000000000000ULL;
@@ -967,7 +985,7 @@ int floatx80_is_signaling_nan(floatx80 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     return 0;
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return ((a.high & 0x7FFF) == 0x7FFF)
             && ((a.low << 1) >= 0x8000000000000000ULL);
     } else {
@@ -991,7 +1009,7 @@ floatx80 floatx80_silence_nan(floatx80 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     g_assert_not_reached();
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return floatx80_default_nan(status);
     } else {
         a.low |= LIT64(0xC000000000000000);
@@ -1105,7 +1123,7 @@ int float128_is_quiet_nan(float128 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     return float128_is_any_nan(a);
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return (((a.high >> 47) & 0xFFFF) == 0xFFFE)
             && (a.low || (a.high & 0x00007FFFFFFFFFFFULL));
     } else {
@@ -1125,7 +1143,7 @@ int float128_is_signaling_nan(float128 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     return 0;
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return ((a.high << 1) >= 0xFFFF000000000000ULL)
             && (a.low || (a.high & 0x0000FFFFFFFFFFFFULL));
     } else {
@@ -1145,7 +1163,7 @@ float128 float128_silence_nan(float128 a, float_status *status)
 #ifdef NO_SIGNALING_NANS
     g_assert_not_reached();
 #else
-    if (status->snan_bit_is_one) {
+    if (snan_bit_is_one(status)) {
         return float128_default_nan(status);
     } else {
         a.high |= LIT64(0x0000800000000000);
diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h
index 4e378cb612..2aae6a89b1 100644
--- a/include/fpu/softfloat-types.h
+++ b/include/fpu/softfloat-types.h
@@ -173,6 +173,7 @@ typedef struct float_status {
     /* should denormalised inputs go to zero and set the input_denormal flag? */
     flag flush_inputs_to_zero;
     flag default_nan_mode;
+    /* not always used -- see snan_bit_is_one() in softfloat-specialize.h */
     flag snan_bit_is_one;
 } float_status;
 
diff --git a/target/hppa/cpu.c b/target/hppa/cpu.c
index c261b6b090..00bf444620 100644
--- a/target/hppa/cpu.c
+++ b/target/hppa/cpu.c
@@ -141,7 +141,6 @@ static void hppa_cpu_initfn(Object *obj)
     cs->env_ptr = env;
     cs->exception_index = -1;
     cpu_hppa_loaded_fr0(env);
-    set_snan_bit_is_one(true, &env->fp_status);
     cpu_hppa_put_psw(env, PSW_W);
 }
 
diff --git a/target/ppc/fpu_helper.c b/target/ppc/fpu_helper.c
index 9ae418a577..d31a933cbb 100644
--- a/target/ppc/fpu_helper.c
+++ b/target/ppc/fpu_helper.c
@@ -3382,7 +3382,6 @@ void helper_xssqrtqp(CPUPPCState *env, uint32_t opcode)
             xt.f128 = xb.f128;
         } else if (float128_is_neg(xb.f128) && !float128_is_zero(xb.f128)) {
             float_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, 1);
-            set_snan_bit_is_one(0, &env->fp_status);
             xt.f128 = float128_default_nan(&env->fp_status);
         }
     }
diff --git a/target/sh4/cpu.c b/target/sh4/cpu.c
index 541ffc2d97..b9f393b7c7 100644
--- a/target/sh4/cpu.c
+++ b/target/sh4/cpu.c
@@ -71,7 +71,6 @@ static void superh_cpu_reset(CPUState *s)
     set_flush_to_zero(1, &env->fp_status);
 #endif
     set_default_nan_mode(1, &env->fp_status);
-    set_snan_bit_is_one(1, &env->fp_status);
 }
 
 static void superh_cpu_disas_set_info(CPUState *cpu, disassemble_info *info)
diff --git a/target/unicore32/cpu.c b/target/unicore32/cpu.c
index 29d160a88d..68f978d80b 100644
--- a/target/unicore32/cpu.c
+++ b/target/unicore32/cpu.c
@@ -70,7 +70,6 @@ static void unicore_ii_cpu_initfn(Object *obj)
 
     set_feature(env, UC32_HWCAP_CMOV);
     set_feature(env, UC32_HWCAP_UCF64);
-    set_snan_bit_is_one(1, &env->ucf64.fp_status);
 }
 
 static void uc32_any_cpu_initfn(Object *obj)
@@ -83,7 +82,6 @@ static void uc32_any_cpu_initfn(Object *obj)
 
     set_feature(env, UC32_HWCAP_CMOV);
     set_feature(env, UC32_HWCAP_UCF64);
-    set_snan_bit_is_one(1, &env->ucf64.fp_status);
 }
 
 static void uc32_cpu_realizefn(DeviceState *dev, Error **errp)
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 22/28] fpu/softfloat: Specialize on snan_bit_is_one

[Peter Maydell]

On 14 May 2018 at 23:12, Richard Henderson <richard.henderson@linaro.org> wrote:
> Only MIPS requires snan_bit_is_one to be variable.  While we are
> specializing softfloat behaviour, allow other targets to eliminate
> this runtime check.
>
> Cc: Aurelien Jarno <aurelien@aurel32.net>
> Cc: Yongbok Kim <yongbok.kim@mips.com>
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Cc: Alexander Graf <agraf@suse.de>
> Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>
> ---
> v5
>   - do not remove the set_snan_bit_is_one function
>   - tidy the language in the snan_bit_is_one block comment

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

I think this change makes the comments in target/riscv/fpu_helper.c
helper_fclass_s() and helper_fclass_d() stale, incidentally.

thanks
-- PMM

Re: [Qemu-devel] [PATCH v5 22/28] fpu/softfloat: Specialize on snan_bit_is_one

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> Only MIPS requires snan_bit_is_one to be variable.  While we are
> specializing softfloat behaviour, allow other targets to eliminate
> this runtime check.
>
> Cc: Aurelien Jarno <aurelien@aurel32.net>
> Cc: Yongbok Kim <yongbok.kim@mips.com>
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Cc: Alexander Graf <agraf@suse.de>
> Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

>
> ---
> v5
>   - do not remove the set_snan_bit_is_one function
>   - tidy the language in the snan_bit_is_one block comment
> ---
>  fpu/softfloat-specialize.h    | 68 ++++++++++++++++++++++-------------
>  include/fpu/softfloat-types.h |  1 +
>  target/hppa/cpu.c             |  1 -
>  target/ppc/fpu_helper.c       |  1 -
>  target/sh4/cpu.c              |  1 -
>  target/unicore32/cpu.c        |  2 --
>  6 files changed, 44 insertions(+), 30 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index d7033b7757..d1e06da75b 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -79,13 +79,31 @@ this code that are retained.
>   * version 2 or later. See the COPYING file in the top-level directory.
>   */
>
> -#if defined(TARGET_XTENSA)
>  /* Define for architectures which deviate from IEEE in not supporting
>   * signaling NaNs (so all NaNs are treated as quiet).
>   */
> +#if defined(TARGET_XTENSA)
>  #define NO_SIGNALING_NANS 1
>  #endif
>
> +/* Define how the architecture discriminates signaling NaNs.
> + * This done with the most significant bit of the fraction.
> + * In IEEE 754-1985 this was implementation defined, but in IEEE 754-2008
> + * the msb must be zero.  MIPS is (so far) unique in supporting both the
> + * 2008 revision and backward compatibility with their original choice.
> + * Thus for MIPS we must make the choice at runtime.
> + */
> +static inline flag snan_bit_is_one(float_status *status)
> +{
> +#if defined(TARGET_MIPS)
> +    return status->snan_bit_is_one;
> +#elif defined(TARGET_HPPA) || defined(TARGET_UNICORE32) || defined(TARGET_SH4)
> +    return 1;
> +#else
> +    return 0;
> +#endif
> +}
> +
>  /*----------------------------------------------------------------------------
>  | For the deconstructed floating-point with fraction FRAC, return true
>  | if the fraction represents a signalling NaN; otherwise false.
> @@ -97,7 +115,7 @@ static bool parts_is_snan_frac(uint64_t frac, float_status *status)
>      return false;
>  #else
>      flag msb = extract64(frac, DECOMPOSED_BINARY_POINT - 1, 1);
> -    return msb == status->snan_bit_is_one;
> +    return msb == snan_bit_is_one(status);
>  #endif
>  }
>
> @@ -118,7 +136,7 @@ static FloatParts parts_default_nan(float_status *status)
>  #elif defined(TARGET_HPPA)
>      frac = 1ULL << (DECOMPOSED_BINARY_POINT - 2);
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          frac = (1ULL << (DECOMPOSED_BINARY_POINT - 1)) - 1;
>      } else {
>  #if defined(TARGET_MIPS)
> @@ -151,7 +169,7 @@ static FloatParts parts_silence_nan(FloatParts a, float_status *status)
>      a.frac &= ~(1ULL << (DECOMPOSED_BINARY_POINT - 1));
>      a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 2);
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return parts_default_nan(status);
>      } else {
>          a.frac |= 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> @@ -169,7 +187,7 @@ float16 float16_default_nan(float_status *status)
>  #if defined(TARGET_ARM)
>      return const_float16(0x7E00);
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return const_float16(0x7DFF);
>      } else {
>  #if defined(TARGET_MIPS)
> @@ -195,7 +213,7 @@ float32 float32_default_nan(float_status *status)
>  #elif defined(TARGET_HPPA)
>      return const_float32(0x7FA00000);
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return const_float32(0x7FBFFFFF);
>      } else {
>  #if defined(TARGET_MIPS)
> @@ -220,7 +238,7 @@ float64 float64_default_nan(float_status *status)
>  #elif defined(TARGET_HPPA)
>      return const_float64(LIT64(0x7FF4000000000000));
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return const_float64(LIT64(0x7FF7FFFFFFFFFFFF));
>      } else {
>  #if defined(TARGET_MIPS)
> @@ -242,7 +260,7 @@ floatx80 floatx80_default_nan(float_status *status)
>      r.low = LIT64(0xFFFFFFFFFFFFFFFF);
>      r.high = 0x7FFF;
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          r.low = LIT64(0xBFFFFFFFFFFFFFFF);
>          r.high = 0x7FFF;
>      } else {
> @@ -274,7 +292,7 @@ float128 float128_default_nan(float_status *status)
>  {
>      float128 r;
>
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          r.low = LIT64(0xFFFFFFFFFFFFFFFF);
>          r.high = LIT64(0x7FFF7FFFFFFFFFFF);
>      } else {
> @@ -319,7 +337,7 @@ int float16_is_quiet_nan(float16 a_, float_status *status)
>      return float16_is_any_nan(a_);
>  #else
>      uint16_t a = float16_val(a_);
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return (((a >> 9) & 0x3F) == 0x3E) && (a & 0x1FF);
>      } else {
>          return ((a & ~0x8000) >= 0x7C80);
> @@ -338,7 +356,7 @@ int float16_is_signaling_nan(float16 a_, float_status *status)
>      return 0;
>  #else
>      uint16_t a = float16_val(a_);
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return ((a & ~0x8000) >= 0x7C80);
>      } else {
>          return (((a >> 9) & 0x3F) == 0x3E) && (a & 0x1FF);
> @@ -356,7 +374,7 @@ float16 float16_silence_nan(float16 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      g_assert_not_reached();
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return float16_default_nan(status);
>      } else {
>          return a | (1 << 9);
> @@ -375,7 +393,7 @@ int float32_is_quiet_nan(float32 a_, float_status *status)
>      return float32_is_any_nan(a_);
>  #else
>      uint32_t a = float32_val(a_);
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return (((a >> 22) & 0x1FF) == 0x1FE) && (a & 0x003FFFFF);
>      } else {
>          return ((uint32_t)(a << 1) >= 0xFF800000);
> @@ -394,7 +412,7 @@ int float32_is_signaling_nan(float32 a_, float_status *status)
>      return 0;
>  #else
>      uint32_t a = float32_val(a_);
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return ((uint32_t)(a << 1) >= 0xFF800000);
>      } else {
>          return (((a >> 22) & 0x1FF) == 0x1FE) && (a & 0x003FFFFF);
> @@ -412,7 +430,7 @@ float32 float32_silence_nan(float32 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      g_assert_not_reached();
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>  # ifdef TARGET_HPPA
>          a &= ~0x00400000;
>          a |=  0x00200000;
> @@ -651,7 +669,7 @@ static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
>          return 3;
>      }
>
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          /* Prefer sNaN over qNaN, in the a, b, c order. */
>          if (aIsSNaN) {
>              return 0;
> @@ -786,7 +804,7 @@ int float64_is_quiet_nan(float64 a_, float_status *status)
>      return float64_is_any_nan(a_);
>  #else
>      uint64_t a = float64_val(a_);
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return (((a >> 51) & 0xFFF) == 0xFFE)
>              && (a & 0x0007FFFFFFFFFFFFULL);
>      } else {
> @@ -806,7 +824,7 @@ int float64_is_signaling_nan(float64 a_, float_status *status)
>      return 0;
>  #else
>      uint64_t a = float64_val(a_);
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return ((a << 1) >= 0xFFF0000000000000ULL);
>      } else {
>          return (((a >> 51) & 0xFFF) == 0xFFE)
> @@ -825,7 +843,7 @@ float64 float64_silence_nan(float64 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      g_assert_not_reached();
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>  # ifdef TARGET_HPPA
>          a &= ~0x0008000000000000ULL;
>          a |=  0x0004000000000000ULL;
> @@ -942,7 +960,7 @@ int floatx80_is_quiet_nan(floatx80 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      return floatx80_is_any_nan(a);
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          uint64_t aLow;
>
>          aLow = a.low & ~0x4000000000000000ULL;
> @@ -967,7 +985,7 @@ int floatx80_is_signaling_nan(floatx80 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      return 0;
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return ((a.high & 0x7FFF) == 0x7FFF)
>              && ((a.low << 1) >= 0x8000000000000000ULL);
>      } else {
> @@ -991,7 +1009,7 @@ floatx80 floatx80_silence_nan(floatx80 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      g_assert_not_reached();
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return floatx80_default_nan(status);
>      } else {
>          a.low |= LIT64(0xC000000000000000);
> @@ -1105,7 +1123,7 @@ int float128_is_quiet_nan(float128 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      return float128_is_any_nan(a);
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return (((a.high >> 47) & 0xFFFF) == 0xFFFE)
>              && (a.low || (a.high & 0x00007FFFFFFFFFFFULL));
>      } else {
> @@ -1125,7 +1143,7 @@ int float128_is_signaling_nan(float128 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      return 0;
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return ((a.high << 1) >= 0xFFFF000000000000ULL)
>              && (a.low || (a.high & 0x0000FFFFFFFFFFFFULL));
>      } else {
> @@ -1145,7 +1163,7 @@ float128 float128_silence_nan(float128 a, float_status *status)
>  #ifdef NO_SIGNALING_NANS
>      g_assert_not_reached();
>  #else
> -    if (status->snan_bit_is_one) {
> +    if (snan_bit_is_one(status)) {
>          return float128_default_nan(status);
>      } else {
>          a.high |= LIT64(0x0000800000000000);
> diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h
> index 4e378cb612..2aae6a89b1 100644
> --- a/include/fpu/softfloat-types.h
> +++ b/include/fpu/softfloat-types.h
> @@ -173,6 +173,7 @@ typedef struct float_status {
>      /* should denormalised inputs go to zero and set the input_denormal flag? */
>      flag flush_inputs_to_zero;
>      flag default_nan_mode;
> +    /* not always used -- see snan_bit_is_one() in softfloat-specialize.h */
>      flag snan_bit_is_one;
>  } float_status;
>
> diff --git a/target/hppa/cpu.c b/target/hppa/cpu.c
> index c261b6b090..00bf444620 100644
> --- a/target/hppa/cpu.c
> +++ b/target/hppa/cpu.c
> @@ -141,7 +141,6 @@ static void hppa_cpu_initfn(Object *obj)
>      cs->env_ptr = env;
>      cs->exception_index = -1;
>      cpu_hppa_loaded_fr0(env);
> -    set_snan_bit_is_one(true, &env->fp_status);
>      cpu_hppa_put_psw(env, PSW_W);
>  }
>
> diff --git a/target/ppc/fpu_helper.c b/target/ppc/fpu_helper.c
> index 9ae418a577..d31a933cbb 100644
> --- a/target/ppc/fpu_helper.c
> +++ b/target/ppc/fpu_helper.c
> @@ -3382,7 +3382,6 @@ void helper_xssqrtqp(CPUPPCState *env, uint32_t opcode)
>              xt.f128 = xb.f128;
>          } else if (float128_is_neg(xb.f128) && !float128_is_zero(xb.f128)) {
>              float_invalid_op_excp(env, POWERPC_EXCP_FP_VXSQRT, 1);
> -            set_snan_bit_is_one(0, &env->fp_status);
>              xt.f128 = float128_default_nan(&env->fp_status);
>          }
>      }
> diff --git a/target/sh4/cpu.c b/target/sh4/cpu.c
> index 541ffc2d97..b9f393b7c7 100644
> --- a/target/sh4/cpu.c
> +++ b/target/sh4/cpu.c
> @@ -71,7 +71,6 @@ static void superh_cpu_reset(CPUState *s)
>      set_flush_to_zero(1, &env->fp_status);
>  #endif
>      set_default_nan_mode(1, &env->fp_status);
> -    set_snan_bit_is_one(1, &env->fp_status);
>  }
>
>  static void superh_cpu_disas_set_info(CPUState *cpu, disassemble_info *info)
> diff --git a/target/unicore32/cpu.c b/target/unicore32/cpu.c
> index 29d160a88d..68f978d80b 100644
> --- a/target/unicore32/cpu.c
> +++ b/target/unicore32/cpu.c
> @@ -70,7 +70,6 @@ static void unicore_ii_cpu_initfn(Object *obj)
>
>      set_feature(env, UC32_HWCAP_CMOV);
>      set_feature(env, UC32_HWCAP_UCF64);
> -    set_snan_bit_is_one(1, &env->ucf64.fp_status);
>  }
>
>  static void uc32_any_cpu_initfn(Object *obj)
> @@ -83,7 +82,6 @@ static void uc32_any_cpu_initfn(Object *obj)
>
>      set_feature(env, UC32_HWCAP_CMOV);
>      set_feature(env, UC32_HWCAP_UCF64);
> -    set_snan_bit_is_one(1, &env->ucf64.fp_status);
>  }
>
>  static void uc32_cpu_realizefn(DeviceState *dev, Error **errp)


--
Alex Bennée

[Qemu-devel] [PATCH v5 23/28] fpu/softfloat: Make is_nan et al available to softfloat-specialize.h

[Richard Henderson]

We will need these helpers within softfloat-specialize.h, so move
the definitions above the include.  After specialization, they will
not always be used so mark them to avoid the Werror.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat.c | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index e9e421728b..9934adbae5 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -181,6 +181,22 @@ typedef enum __attribute__ ((__packed__)) {
     float_class_snan,
 } FloatClass;
 
+/* Simple helpers for checking if, or what kind of, NaN we have */
+static inline __attribute__((unused)) bool is_nan(FloatClass c)
+{
+    return unlikely(c >= float_class_qnan);
+}
+
+static inline __attribute__((unused)) bool is_snan(FloatClass c)
+{
+    return c == float_class_snan;
+}
+
+static inline __attribute__((unused)) bool is_qnan(FloatClass c)
+{
+    return c == float_class_qnan;
+}
+
 /*
  * Structure holding all of the decomposed parts of a float. The
  * exponent is unbiased and the fraction is normalized. All
@@ -536,20 +552,6 @@ static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
     return float64_pack_raw(round_canonical(p, s, &float64_params));
 }
 
-/* Simple helpers for checking if what NaN we have */
-static bool is_nan(FloatClass c)
-{
-    return unlikely(c >= float_class_qnan);
-}
-static bool is_snan(FloatClass c)
-{
-    return c == float_class_snan;
-}
-static bool is_qnan(FloatClass c)
-{
-    return c == float_class_qnan;
-}
-
 static FloatParts return_nan(FloatParts a, float_status *s)
 {
     switch (a.cls) {
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 23/28] fpu/softfloat: Make is_nan et al available to softfloat-specialize.h

[Peter Maydell]

On 14 May 2018 at 23:12, Richard Henderson <richard.henderson@linaro.org> wrote:
> We will need these helpers within softfloat-specialize.h, so move
> the definitions above the include.  After specialization, they will
> not always be used so mark them to avoid the Werror.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  fpu/softfloat.c | 30 ++++++++++++++++--------------
>  1 file changed, 16 insertions(+), 14 deletions(-)

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

thanks
-- PMM

Re: [Qemu-devel] [PATCH v5 23/28] fpu/softfloat: Make is_nan et al available to softfloat-specialize.h

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> We will need these helpers within softfloat-specialize.h, so move
> the definitions above the include.  After specialization, they will
> not always be used so mark them to avoid the Werror.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat.c | 30 ++++++++++++++++--------------
>  1 file changed, 16 insertions(+), 14 deletions(-)
>
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index e9e421728b..9934adbae5 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -181,6 +181,22 @@ typedef enum __attribute__ ((__packed__)) {
>      float_class_snan,
>  } FloatClass;
>
> +/* Simple helpers for checking if, or what kind of, NaN we have */
> +static inline __attribute__((unused)) bool is_nan(FloatClass c)
> +{
> +    return unlikely(c >= float_class_qnan);
> +}
> +
> +static inline __attribute__((unused)) bool is_snan(FloatClass c)
> +{
> +    return c == float_class_snan;
> +}
> +
> +static inline __attribute__((unused)) bool is_qnan(FloatClass c)
> +{
> +    return c == float_class_qnan;
> +}
> +
>  /*
>   * Structure holding all of the decomposed parts of a float. The
>   * exponent is unbiased and the fraction is normalized. All
> @@ -536,20 +552,6 @@ static float64 float64_round_pack_canonical(FloatParts p, float_status *s)
>      return float64_pack_raw(round_canonical(p, s, &float64_params));
>  }
>
> -/* Simple helpers for checking if what NaN we have */
> -static bool is_nan(FloatClass c)
> -{
> -    return unlikely(c >= float_class_qnan);
> -}
> -static bool is_snan(FloatClass c)
> -{
> -    return c == float_class_snan;
> -}
> -static bool is_qnan(FloatClass c)
> -{
> -    return c == float_class_qnan;
> -}
> -
>  static FloatParts return_nan(FloatParts a, float_status *s)
>  {
>      switch (a.cls) {


--
Alex Bennée

[Qemu-devel] [PATCH v5 24/28] fpu/softfloat: Pass FloatClass to pickNaN

[Richard Henderson]

For each operand, pass a single enumeration instead of a pair of booleans.
The commit also merges multiple different ifdef-selected implementations
of pickNaN into a single function whose body is ifdef-selected.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 168 ++++++++++++++++++-------------------
 fpu/softfloat.c            |   3 +-
 2 files changed, 82 insertions(+), 89 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index d1e06da75b..2695183188 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -501,10 +501,10 @@ static float32 commonNaNToFloat32(commonNaNT a, float_status *status)
 | tie-break rule.
 *----------------------------------------------------------------------------*/
 
-#if defined(TARGET_ARM)
-static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
+static int pickNaN(FloatClass a_cls, FloatClass b_cls,
                    flag aIsLargerSignificand)
 {
+#if defined(TARGET_ARM) || defined(TARGET_MIPS) || defined(TARGET_HPPA)
     /* ARM mandated NaN propagation rules (see FPProcessNaNs()), take
      * the first of:
      *  1. A if it is signaling
@@ -513,20 +513,6 @@ static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
      *  4. B (quiet)
      * A signaling NaN is always quietened before returning it.
      */
-    if (aIsSNaN) {
-        return 0;
-    } else if (bIsSNaN) {
-        return 1;
-    } else if (aIsQNaN) {
-        return 0;
-    } else {
-        return 1;
-    }
-}
-#elif defined(TARGET_MIPS) || defined(TARGET_HPPA)
-static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                    flag aIsLargerSignificand)
-{
     /* According to MIPS specifications, if one of the two operands is
      * a sNaN, a new qNaN has to be generated. This is done in
      * floatXX_silence_nan(). For qNaN inputs the specifications
@@ -540,35 +526,21 @@ static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
      *  4. B (quiet)
      * A signaling NaN is always silenced before returning it.
      */
-    if (aIsSNaN) {
+    if (is_snan(a_cls)) {
         return 0;
-    } else if (bIsSNaN) {
+    } else if (is_snan(b_cls)) {
         return 1;
-    } else if (aIsQNaN) {
+    } else if (is_qnan(a_cls)) {
         return 0;
     } else {
         return 1;
     }
-}
-#elif defined(TARGET_PPC) || defined(TARGET_XTENSA)
-static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                   flag aIsLargerSignificand)
-{
+#elif defined(TARGET_PPC) || defined(TARGET_XTENSA) || defined(TARGET_M68K)
     /* PowerPC propagation rules:
      *  1. A if it sNaN or qNaN
      *  2. B if it sNaN or qNaN
      * A signaling NaN is always silenced before returning it.
      */
-    if (aIsSNaN || aIsQNaN) {
-        return 0;
-    } else {
-        return 1;
-    }
-}
-#elif defined(TARGET_M68K)
-static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                   flag aIsLargerSignificand)
-{
     /* M68000 FAMILY PROGRAMMER'S REFERENCE MANUAL
      * 3.4 FLOATING-POINT INSTRUCTION DETAILS
      * If either operand, but not both operands, of an operation is a
@@ -583,16 +555,12 @@ static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
      * a nonsignaling NaN. The operation then continues as described in the
      * preceding paragraph for nonsignaling NaNs.
      */
-    if (aIsQNaN || aIsSNaN) { /* a is the destination operand */
-        return 0; /* return the destination operand */
+    if (is_nan(a_cls)) {
+        return 0;
     } else {
-        return 1; /* return b */
+        return 1;
     }
-}
 #else
-static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                    flag aIsLargerSignificand)
-{
     /* This implements x87 NaN propagation rules:
      * SNaN + QNaN => return the QNaN
      * two SNaNs => return the one with the larger significand, silenced
@@ -603,13 +571,13 @@ static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
      * If we get down to comparing significands and they are the same,
      * return the NaN with the positive sign bit (if any).
      */
-    if (aIsSNaN) {
-        if (bIsSNaN) {
+    if (is_snan(a_cls)) {
+        if (is_snan(b_cls)) {
             return aIsLargerSignificand ? 0 : 1;
         }
-        return bIsQNaN ? 1 : 0;
-    } else if (aIsQNaN) {
-        if (bIsSNaN || !bIsQNaN) {
+        return is_qnan(b_cls) ? 1 : 0;
+    } else if (is_qnan(a_cls)) {
+        if (is_snan(b_cls) || !is_qnan(b_cls)) {
             return 0;
         } else {
             return aIsLargerSignificand ? 0 : 1;
@@ -617,8 +585,8 @@ static int pickNaN(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
     } else {
         return 1;
     }
-}
 #endif
+}
 
 /*----------------------------------------------------------------------------
 | Select which NaN to propagate for a three-input operation.
@@ -752,18 +720,26 @@ static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
 
 static float32 propagateFloat32NaN(float32 a, float32 b, float_status *status)
 {
-    flag aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN;
     flag aIsLargerSignificand;
     uint32_t av, bv;
+    FloatClass a_cls, b_cls;
+
+    /* This is not complete, but is good enough for pickNaN.  */
+    a_cls = (!float32_is_any_nan(a)
+             ? float_class_normal
+             : float32_is_signaling_nan(a, status)
+             ? float_class_snan
+             : float_class_qnan);
+    b_cls = (!float32_is_any_nan(b)
+             ? float_class_normal
+             : float32_is_signaling_nan(b, status)
+             ? float_class_snan
+             : float_class_qnan);
 
-    aIsQuietNaN = float32_is_quiet_nan(a, status);
-    aIsSignalingNaN = float32_is_signaling_nan(a, status);
-    bIsQuietNaN = float32_is_quiet_nan(b, status);
-    bIsSignalingNaN = float32_is_signaling_nan(b, status);
     av = float32_val(a);
     bv = float32_val(b);
 
-    if (aIsSignalingNaN | bIsSignalingNaN) {
+    if (is_snan(a_cls) || is_snan(b_cls)) {
         float_raise(float_flag_invalid, status);
     }
 
@@ -779,14 +755,13 @@ static float32 propagateFloat32NaN(float32 a, float32 b, float_status *status)
         aIsLargerSignificand = (av < bv) ? 1 : 0;
     }
 
-    if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
-                aIsLargerSignificand)) {
-        if (bIsSignalingNaN) {
+    if (pickNaN(a_cls, b_cls, aIsLargerSignificand)) {
+        if (is_snan(b_cls)) {
             return float32_silence_nan(b, status);
         }
         return b;
     } else {
-        if (aIsSignalingNaN) {
+        if (is_snan(a_cls)) {
             return float32_silence_nan(a, status);
         }
         return a;
@@ -908,18 +883,26 @@ static float64 commonNaNToFloat64(commonNaNT a, float_status *status)
 
 static float64 propagateFloat64NaN(float64 a, float64 b, float_status *status)
 {
-    flag aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN;
     flag aIsLargerSignificand;
     uint64_t av, bv;
+    FloatClass a_cls, b_cls;
+
+    /* This is not complete, but is good enough for pickNaN.  */
+    a_cls = (!float64_is_any_nan(a)
+             ? float_class_normal
+             : float64_is_signaling_nan(a, status)
+             ? float_class_snan
+             : float_class_qnan);
+    b_cls = (!float64_is_any_nan(b)
+             ? float_class_normal
+             : float64_is_signaling_nan(b, status)
+             ? float_class_snan
+             : float_class_qnan);
 
-    aIsQuietNaN = float64_is_quiet_nan(a, status);
-    aIsSignalingNaN = float64_is_signaling_nan(a, status);
-    bIsQuietNaN = float64_is_quiet_nan(b, status);
-    bIsSignalingNaN = float64_is_signaling_nan(b, status);
     av = float64_val(a);
     bv = float64_val(b);
 
-    if (aIsSignalingNaN | bIsSignalingNaN) {
+    if (is_snan(a_cls) || is_snan(b_cls)) {
         float_raise(float_flag_invalid, status);
     }
 
@@ -935,14 +918,13 @@ static float64 propagateFloat64NaN(float64 a, float64 b, float_status *status)
         aIsLargerSignificand = (av < bv) ? 1 : 0;
     }
 
-    if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
-                aIsLargerSignificand)) {
-        if (bIsSignalingNaN) {
+    if (pickNaN(a_cls, b_cls, aIsLargerSignificand)) {
+        if (is_snan(b_cls)) {
             return float64_silence_nan(b, status);
         }
         return b;
     } else {
-        if (aIsSignalingNaN) {
+        if (is_snan(a_cls)) {
             return float64_silence_nan(a, status);
         }
         return a;
@@ -1075,15 +1057,22 @@ static floatx80 commonNaNToFloatx80(commonNaNT a, float_status *status)
 
 floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status)
 {
-    flag aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN;
     flag aIsLargerSignificand;
+    FloatClass a_cls, b_cls;
 
-    aIsQuietNaN = floatx80_is_quiet_nan(a, status);
-    aIsSignalingNaN = floatx80_is_signaling_nan(a, status);
-    bIsQuietNaN = floatx80_is_quiet_nan(b, status);
-    bIsSignalingNaN = floatx80_is_signaling_nan(b, status);
+    /* This is not complete, but is good enough for pickNaN.  */
+    a_cls = (!floatx80_is_any_nan(a)
+             ? float_class_normal
+             : floatx80_is_signaling_nan(a, status)
+             ? float_class_snan
+             : float_class_qnan);
+    b_cls = (!floatx80_is_any_nan(b)
+             ? float_class_normal
+             : floatx80_is_signaling_nan(b, status)
+             ? float_class_snan
+             : float_class_qnan);
 
-    if (aIsSignalingNaN | bIsSignalingNaN) {
+    if (is_snan(a_cls) || is_snan(b_cls)) {
         float_raise(float_flag_invalid, status);
     }
 
@@ -1099,14 +1088,13 @@ floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status)
         aIsLargerSignificand = (a.high < b.high) ? 1 : 0;
     }
 
-    if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
-                aIsLargerSignificand)) {
-        if (bIsSignalingNaN) {
+    if (pickNaN(a_cls, b_cls, aIsLargerSignificand)) {
+        if (is_snan(b_cls)) {
             return floatx80_silence_nan(b, status);
         }
         return b;
     } else {
-        if (aIsSignalingNaN) {
+        if (is_snan(a_cls)) {
             return floatx80_silence_nan(a, status);
         }
         return a;
@@ -1217,15 +1205,22 @@ static float128 commonNaNToFloat128(commonNaNT a, float_status *status)
 static float128 propagateFloat128NaN(float128 a, float128 b,
                                      float_status *status)
 {
-    flag aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN;
     flag aIsLargerSignificand;
+    FloatClass a_cls, b_cls;
 
-    aIsQuietNaN = float128_is_quiet_nan(a, status);
-    aIsSignalingNaN = float128_is_signaling_nan(a, status);
-    bIsQuietNaN = float128_is_quiet_nan(b, status);
-    bIsSignalingNaN = float128_is_signaling_nan(b, status);
+    /* This is not complete, but is good enough for pickNaN.  */
+    a_cls = (!float128_is_any_nan(a)
+             ? float_class_normal
+             : float128_is_signaling_nan(a, status)
+             ? float_class_snan
+             : float_class_qnan);
+    b_cls = (!float128_is_any_nan(b)
+             ? float_class_normal
+             : float128_is_signaling_nan(b, status)
+             ? float_class_snan
+             : float_class_qnan);
 
-    if (aIsSignalingNaN | bIsSignalingNaN) {
+    if (is_snan(a_cls) || is_snan(b_cls)) {
         float_raise(float_flag_invalid, status);
     }
 
@@ -1241,14 +1236,13 @@ static float128 propagateFloat128NaN(float128 a, float128 b,
         aIsLargerSignificand = (a.high < b.high) ? 1 : 0;
     }
 
-    if (pickNaN(aIsQuietNaN, aIsSignalingNaN, bIsQuietNaN, bIsSignalingNaN,
-                aIsLargerSignificand)) {
-        if (bIsSignalingNaN) {
+    if (pickNaN(a_cls, b_cls, aIsLargerSignificand)) {
+        if (is_snan(b_cls)) {
             return float128_silence_nan(b, status);
         }
         return b;
     } else {
-        if (aIsSignalingNaN) {
+        if (is_snan(a_cls)) {
             return float128_silence_nan(a, status);
         }
         return a;
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 9934adbae5..43264fa1f6 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -580,8 +580,7 @@ static FloatParts pick_nan(FloatParts a, FloatParts b, float_status *s)
     if (s->default_nan_mode) {
         return parts_default_nan(s);
     } else {
-        if (pickNaN(is_qnan(a.cls), is_snan(a.cls),
-                    is_qnan(b.cls), is_snan(b.cls),
+        if (pickNaN(a.cls, b.cls,
                     a.frac > b.frac ||
                     (a.frac == b.frac && a.sign < b.sign))) {
             a = b;
-- 
2.17.0

[Qemu-devel] [PATCH v5 25/28] fpu/softfloat: Pass FloatClass to pickNaNMulAdd

[Richard Henderson]

For each operand, pass a single enumeration instead of a pair of booleans.
The commit also merges multiple different ifdef-selected implementations
of pickNaNMulAdd into a single function whose body is ifdef-selected.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 70 +++++++++++++++-----------------------
 fpu/softfloat.c            |  5 +--
 2 files changed, 28 insertions(+), 47 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 2695183188..0399dfe011 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -594,15 +594,14 @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls,
 | information.
 | Return values : 0 : a; 1 : b; 2 : c; 3 : default-NaN
 *----------------------------------------------------------------------------*/
-#if defined(TARGET_ARM)
-static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                         flag cIsQNaN, flag cIsSNaN, flag infzero,
-                         float_status *status)
+static int pickNaNMulAdd(FloatClass a_cls, FloatClass b_cls, FloatClass c_cls,
+                         bool infzero, float_status *status)
 {
+#if defined(TARGET_ARM)
     /* For ARM, the (inf,zero,qnan) case sets InvalidOp and returns
      * the default NaN
      */
-    if (infzero && cIsQNaN) {
+    if (infzero && is_qnan(c_cls)) {
         float_raise(float_flag_invalid, status);
         return 3;
     }
@@ -610,25 +609,20 @@ static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
     /* 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.
      */
-    if (cIsSNaN) {
+    if (is_snan(c_cls)) {
         return 2;
-    } else if (aIsSNaN) {
+    } else if (is_snan(a_cls)) {
         return 0;
-    } else if (bIsSNaN) {
+    } else if (is_snan(b_cls)) {
         return 1;
-    } else if (cIsQNaN) {
+    } else if (is_qnan(c_cls)) {
         return 2;
-    } else if (aIsQNaN) {
+    } else if (is_qnan(a_cls)) {
         return 0;
     } else {
         return 1;
     }
-}
 #elif defined(TARGET_MIPS)
-static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                         flag cIsQNaN, flag cIsSNaN, flag infzero,
-                         float_status *status)
-{
     /* For MIPS, the (inf,zero,qnan) case sets InvalidOp and returns
      * the default NaN
      */
@@ -639,41 +633,36 @@ static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
 
     if (snan_bit_is_one(status)) {
         /* Prefer sNaN over qNaN, in the a, b, c order. */
-        if (aIsSNaN) {
+        if (is_snan(a_cls)) {
             return 0;
-        } else if (bIsSNaN) {
+        } else if (is_snan(b_cls)) {
             return 1;
-        } else if (cIsSNaN) {
+        } else if (is_snan(c_cls)) {
             return 2;
-        } else if (aIsQNaN) {
+        } else if (is_qnan(a_cls)) {
             return 0;
-        } else if (bIsQNaN) {
+        } else if (is_qnan(b_cls)) {
             return 1;
         } else {
             return 2;
         }
     } else {
         /* Prefer sNaN over qNaN, in the c, a, b order. */
-        if (cIsSNaN) {
+        if (is_snan(c_cls)) {
             return 2;
-        } else if (aIsSNaN) {
+        } else if (is_snan(a_cls)) {
             return 0;
-        } else if (bIsSNaN) {
+        } else if (is_snan(b_cls)) {
             return 1;
-        } else if (cIsQNaN) {
+        } else if (is_qnan(c_cls)) {
             return 2;
-        } else if (aIsQNaN) {
+        } else if (is_qnan(a_cls)) {
             return 0;
         } else {
             return 1;
         }
     }
-}
 #elif defined(TARGET_PPC)
-static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                         flag cIsQNaN, flag cIsSNaN, flag infzero,
-                         float_status *status)
-{
     /* 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
@@ -686,31 +675,26 @@ static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
     /* 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
      */
-    if (aIsSNaN || aIsQNaN) {
+    if (is_nan(a_cls)) {
         return 0;
-    } else if (cIsSNaN || cIsQNaN) {
+    } else if (is_nan(c_cls)) {
         return 2;
     } else {
         return 1;
     }
-}
 #else
-/* A default implementation: prefer a to b to c.
- * This is unlikely to actually match any real implementation.
- */
-static int pickNaNMulAdd(flag aIsQNaN, flag aIsSNaN, flag bIsQNaN, flag bIsSNaN,
-                         flag cIsQNaN, flag cIsSNaN, flag infzero,
-                         float_status *status)
-{
-    if (aIsSNaN || aIsQNaN) {
+    /* A default implementation: prefer a to b to c.
+     * This is unlikely to actually match any real implementation.
+     */
+    if (is_nan(a_cls)) {
         return 0;
-    } else if (bIsSNaN || bIsQNaN) {
+    } else if (is_nan(b_cls)) {
         return 1;
     } else {
         return 2;
     }
-}
 #endif
+}
 
 /*----------------------------------------------------------------------------
 | Takes two single-precision floating-point values `a' and `b', one of which
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 43264fa1f6..65e970eeb2 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -601,10 +601,7 @@ static FloatParts pick_nan_muladd(FloatParts a, FloatParts b, FloatParts c,
         s->float_exception_flags |= float_flag_invalid;
     }
 
-    which = pickNaNMulAdd(is_qnan(a.cls), is_snan(a.cls),
-                          is_qnan(b.cls), is_snan(b.cls),
-                          is_qnan(c.cls), is_snan(c.cls),
-                          inf_zero, s);
+    which = pickNaNMulAdd(a.cls, b.cls, c.cls, inf_zero, s);
 
     if (s->default_nan_mode) {
         /* Note that this check is after pickNaNMulAdd so that function
-- 
2.17.0

[Qemu-devel] [PATCH v5 26/28] fpu/softfloat: Define floatN_default_nan in terms of parts_default_nan

[Richard Henderson]

Isolate the target-specific choice to 2 functions instead of 6.

The code in float16_default_nan was only correct for ARM, MIPS, and X86.
Though float16 support is rare among our targets.

The code in float128_default_nan was arguably wrong for Sparc.  While
QEMU supports the Sparc 128-bit insns, no real cpu enables it.

The code in floatx80_default_nan tried to be over-general.  There are
only two targets that support this format: x86 and m68k.  Thus there
is no point in inventing a value for snan_bit_is_one.

Move routines that no longer have ifdefs out of softfloat-specialize.h.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 105 +++----------------------------------
 fpu/softfloat.c            |  35 +++++++++++++
 2 files changed, 41 insertions(+), 99 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 0399dfe011..9d562ed504 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -179,94 +179,22 @@ static FloatParts parts_silence_nan(FloatParts a, float_status *status)
     return a;
 }
 
-/*----------------------------------------------------------------------------
-| The pattern for a default generated half-precision NaN.
-*----------------------------------------------------------------------------*/
-float16 float16_default_nan(float_status *status)
-{
-#if defined(TARGET_ARM)
-    return const_float16(0x7E00);
-#else
-    if (snan_bit_is_one(status)) {
-        return const_float16(0x7DFF);
-    } else {
-#if defined(TARGET_MIPS)
-        return const_float16(0x7E00);
-#else
-        return const_float16(0xFE00);
-#endif
-    }
-#endif
-}
-
-/*----------------------------------------------------------------------------
-| The pattern for a default generated single-precision NaN.
-*----------------------------------------------------------------------------*/
-float32 float32_default_nan(float_status *status)
-{
-#if defined(TARGET_SPARC) || defined(TARGET_M68K)
-    return const_float32(0x7FFFFFFF);
-#elif defined(TARGET_PPC) || defined(TARGET_ARM) || defined(TARGET_ALPHA) || \
-      defined(TARGET_XTENSA) || defined(TARGET_S390X) || \
-      defined(TARGET_TRICORE) || defined(TARGET_RISCV)
-    return const_float32(0x7FC00000);
-#elif defined(TARGET_HPPA)
-    return const_float32(0x7FA00000);
-#else
-    if (snan_bit_is_one(status)) {
-        return const_float32(0x7FBFFFFF);
-    } else {
-#if defined(TARGET_MIPS)
-        return const_float32(0x7FC00000);
-#else
-        return const_float32(0xFFC00000);
-#endif
-    }
-#endif
-}
-
-/*----------------------------------------------------------------------------
-| The pattern for a default generated double-precision NaN.
-*----------------------------------------------------------------------------*/
-float64 float64_default_nan(float_status *status)
-{
-#if defined(TARGET_SPARC) || defined(TARGET_M68K)
-    return const_float64(LIT64(0x7FFFFFFFFFFFFFFF));
-#elif defined(TARGET_PPC) || defined(TARGET_ARM) || defined(TARGET_ALPHA) || \
-      defined(TARGET_S390X) || defined(TARGET_RISCV)
-    return const_float64(LIT64(0x7FF8000000000000));
-#elif defined(TARGET_HPPA)
-    return const_float64(LIT64(0x7FF4000000000000));
-#else
-    if (snan_bit_is_one(status)) {
-        return const_float64(LIT64(0x7FF7FFFFFFFFFFFF));
-    } else {
-#if defined(TARGET_MIPS)
-        return const_float64(LIT64(0x7FF8000000000000));
-#else
-        return const_float64(LIT64(0xFFF8000000000000));
-#endif
-    }
-#endif
-}
-
 /*----------------------------------------------------------------------------
 | The pattern for a default generated extended double-precision NaN.
 *----------------------------------------------------------------------------*/
 floatx80 floatx80_default_nan(float_status *status)
 {
     floatx80 r;
+
+    /* None of the targets that have snan_bit_is_one use floatx80.  */
+    assert(!snan_bit_is_one(status));
 #if defined(TARGET_M68K)
     r.low = LIT64(0xFFFFFFFFFFFFFFFF);
     r.high = 0x7FFF;
 #else
-    if (snan_bit_is_one(status)) {
-        r.low = LIT64(0xBFFFFFFFFFFFFFFF);
-        r.high = 0x7FFF;
-    } else {
-        r.low = LIT64(0xC000000000000000);
-        r.high = 0xFFFF;
-    }
+    /* X86 */
+    r.low = LIT64(0xC000000000000000);
+    r.high = 0xFFFF;
 #endif
     return r;
 }
@@ -285,27 +213,6 @@ floatx80 floatx80_default_nan(float_status *status)
 const floatx80 floatx80_infinity
     = make_floatx80_init(floatx80_infinity_high, floatx80_infinity_low);
 
-/*----------------------------------------------------------------------------
-| The pattern for a default generated quadruple-precision NaN.
-*----------------------------------------------------------------------------*/
-float128 float128_default_nan(float_status *status)
-{
-    float128 r;
-
-    if (snan_bit_is_one(status)) {
-        r.low = LIT64(0xFFFFFFFFFFFFFFFF);
-        r.high = LIT64(0x7FFF7FFFFFFFFFFF);
-    } else {
-        r.low = LIT64(0x0000000000000000);
-#if defined(TARGET_S390X) || defined(TARGET_PPC) || defined(TARGET_RISCV)
-        r.high = LIT64(0x7FFF800000000000);
-#else
-        r.high = LIT64(0xFFFF800000000000);
-#endif
-    }
-    return r;
-}
-
 /*----------------------------------------------------------------------------
 | Raises the exceptions specified by `flags'.  Floating-point traps can be
 | defined here if desired.  It is currently not possible for such a trap
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 65e970eeb2..b5842f7b1c 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -2092,6 +2092,41 @@ float64 __attribute__((flatten)) float64_sqrt(float64 a, float_status *status)
     return float64_round_pack_canonical(pr, status);
 }
 
+/*----------------------------------------------------------------------------
+| The pattern for a default generated NaN.
+*----------------------------------------------------------------------------*/
+
+float16 float16_default_nan(float_status *status)
+{
+    return float16_pack_raw(parts_default_nan(status));
+}
+
+float32 float32_default_nan(float_status *status)
+{
+    return float32_pack_raw(parts_default_nan(status));
+}
+
+float64 float64_default_nan(float_status *status)
+{
+    return float64_pack_raw(parts_default_nan(status));
+}
+
+float128 float128_default_nan(float_status *status)
+{
+    FloatParts p = parts_default_nan(status);
+    float128 r;
+
+    /* Extrapolate from the choices made by parts_default_nan to fill
+     * in the quad-floating format.  If the low bit is set, assume we
+     * want to set all non-snan bits.
+     */
+    r.low = -(p.frac & 1);
+    r.high = p.frac >> (DECOMPOSED_BINARY_POINT - 48);
+    r.high |= LIT64(0x7FFF000000000000);
+    r.high |= (uint64_t)p.sign << 63;
+
+    return r;
+}
 
 /*----------------------------------------------------------------------------
 | Takes a 64-bit fixed-point value `absZ' with binary point between bits 6
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 26/28] fpu/softfloat: Define floatN_default_nan in terms of parts_default_nan

[Peter Maydell]

On 14 May 2018 at 23:12, Richard Henderson <richard.henderson@linaro.org> wrote:
> Isolate the target-specific choice to 2 functions instead of 6.
>
> The code in float16_default_nan was only correct for ARM, MIPS, and X86.
> Though float16 support is rare among our targets.
>
> The code in float128_default_nan was arguably wrong for Sparc.  While
> QEMU supports the Sparc 128-bit insns, no real cpu enables it.
>
> The code in floatx80_default_nan tried to be over-general.  There are
> only two targets that support this format: x86 and m68k.  Thus there
> is no point in inventing a value for snan_bit_is_one.
>
> Move routines that no longer have ifdefs out of softfloat-specialize.h.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---

This patch results in float32_default_nan() for Arm returning 0x7f800000,
which is not a NaN...

thanks
-- PMM

[Qemu-devel] [PATCH v5 27/28] fpu/softfloat: Clean up parts_default_nan

[Richard Henderson]

Reduce the number of ifdefs.  Correct the result for OpenRISC
and TriCore (although TriCore fixed in target-specific code).

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 21 ++++++++++++++-------
 1 file changed, 14 insertions(+), 7 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index 9d562ed504..ec4fb6ba8b 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -129,22 +129,29 @@ static FloatParts parts_default_nan(float_status *status)
     uint64_t frac;
 
 #if defined(TARGET_SPARC) || defined(TARGET_M68K)
+    /* !snan_bit_is_one, set all bits */
     frac = (1ULL << DECOMPOSED_BINARY_POINT) - 1;
-#elif defined(TARGET_PPC) || defined(TARGET_ARM) || defined(TARGET_ALPHA) || \
-      defined(TARGET_S390X) || defined(TARGET_RISCV)
+#elif defined(TARGET_I386) || defined(TARGET_X86_64) \
+    || defined(TARGET_MICROBLAZE)
+    /* !snan_bit_is_one, set sign and msb */
     frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
+    sign = 1;
 #elif defined(TARGET_HPPA)
+    /* snan_bit_is_one, set msb-1.  */
     frac = 1ULL << (DECOMPOSED_BINARY_POINT - 2);
 #else
+    /* This case is true for Alpha, ARM, MIPS, OpenRISC, PPC, RISC-V,
+     * S390, SH4, TriCore, and Xtensa.  I cannot find documentation
+     * for Unicore32; the choice from the original commit is unchanged.
+     * Our other supported targets, CRIS, LM32, Moxie, Nios2, and Tile,
+     * do not have floating-point.
+     */
     if (snan_bit_is_one(status)) {
+        /* set all bits other than msb */
         frac = (1ULL << (DECOMPOSED_BINARY_POINT - 1)) - 1;
     } else {
-#if defined(TARGET_MIPS)
+        /* set msb */
         frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
-#else
-        frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
-        sign = 1;
-#endif
     }
 #endif
 
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 27/28] fpu/softfloat: Clean up parts_default_nan

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> Reduce the number of ifdefs.  Correct the result for OpenRISC
> and TriCore (although TriCore fixed in target-specific code).
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  fpu/softfloat-specialize.h | 21 ++++++++++++++-------
>  1 file changed, 14 insertions(+), 7 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index 9d562ed504..ec4fb6ba8b 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -129,22 +129,29 @@ static FloatParts parts_default_nan(float_status *status)
>      uint64_t frac;
>
>  #if defined(TARGET_SPARC) || defined(TARGET_M68K)
> +    /* !snan_bit_is_one, set all bits */
>      frac = (1ULL << DECOMPOSED_BINARY_POINT) - 1;
> -#elif defined(TARGET_PPC) || defined(TARGET_ARM) || defined(TARGET_ALPHA) || \
> -      defined(TARGET_S390X) || defined(TARGET_RISCV)
> +#elif defined(TARGET_I386) || defined(TARGET_X86_64) \
> +    || defined(TARGET_MICROBLAZE)
> +    /* !snan_bit_is_one, set sign and msb */
>      frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> +    sign = 1;
>  #elif defined(TARGET_HPPA)
> +    /* snan_bit_is_one, set msb-1.  */
>      frac = 1ULL << (DECOMPOSED_BINARY_POINT - 2);
>  #else
> +    /* This case is true for Alpha, ARM, MIPS, OpenRISC, PPC, RISC-V,
> +     * S390, SH4, TriCore, and Xtensa.  I cannot find documentation
> +     * for Unicore32; the choice from the original commit is unchanged.
> +     * Our other supported targets, CRIS, LM32, Moxie, Nios2, and Tile,
> +     * do not have floating-point.
> +     */
>      if (snan_bit_is_one(status)) {
> +        /* set all bits other than msb */
>          frac = (1ULL << (DECOMPOSED_BINARY_POINT - 1)) - 1;
>      } else {
> -#if defined(TARGET_MIPS)
> +        /* set msb */
>          frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> -#else
> -        frac = 1ULL << (DECOMPOSED_BINARY_POINT - 1);
> -        sign = 1;
> -#endif
>      }
>  #endif


--
Alex Bennée

[Qemu-devel] [PATCH v5 28/28] fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan

[Richard Henderson]

Isolate the target-specific choice to 3 functions instead of 6.

The code in floatx80_default_nan tried to be over-general.  There are
only two targets that support this format: x86 and m68k.  Thus there
is no point in inventing a mechanism for snan_bit_is_one.

Move routines that no longer have ifdefs out of softfloat-specialize.h.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 fpu/softfloat-specialize.h | 81 ++------------------------------------
 fpu/softfloat.c            | 19 +++++++++
 2 files changed, 23 insertions(+), 77 deletions(-)

diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
index ec4fb6ba8b..16c0bcb6fa 100644
--- a/fpu/softfloat-specialize.h
+++ b/fpu/softfloat-specialize.h
@@ -278,24 +278,6 @@ int float16_is_signaling_nan(float16 a_, float_status *status)
 #endif
 }
 
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN from a signalling NaN for the half-precision
-| floating point value `a'.
-*----------------------------------------------------------------------------*/
-
-float16 float16_silence_nan(float16 a, float_status *status)
-{
-#ifdef NO_SIGNALING_NANS
-    g_assert_not_reached();
-#else
-    if (snan_bit_is_one(status)) {
-        return float16_default_nan(status);
-    } else {
-        return a | (1 << 9);
-    }
-#endif
-}
-
 /*----------------------------------------------------------------------------
 | Returns 1 if the single-precision floating-point value `a' is a quiet
 | NaN; otherwise returns 0.
@@ -334,30 +316,6 @@ int float32_is_signaling_nan(float32 a_, float_status *status)
 #endif
 }
 
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN from a signalling NaN for the single-precision
-| floating point value `a'.
-*----------------------------------------------------------------------------*/
-
-float32 float32_silence_nan(float32 a, float_status *status)
-{
-#ifdef NO_SIGNALING_NANS
-    g_assert_not_reached();
-#else
-    if (snan_bit_is_one(status)) {
-# ifdef TARGET_HPPA
-        a &= ~0x00400000;
-        a |=  0x00200000;
-        return a;
-# else
-        return float32_default_nan(status);
-# endif
-    } else {
-        return a | (1 << 22);
-    }
-#endif
-}
-
 /*----------------------------------------------------------------------------
 | Returns the result of converting the single-precision floating-point NaN
 | `a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
@@ -706,31 +664,6 @@ int float64_is_signaling_nan(float64 a_, float_status *status)
 #endif
 }
 
-/*----------------------------------------------------------------------------
-| Returns a quiet NaN from a signalling NaN for the double-precision
-| floating point value `a'.
-*----------------------------------------------------------------------------*/
-
-float64 float64_silence_nan(float64 a, float_status *status)
-{
-#ifdef NO_SIGNALING_NANS
-    g_assert_not_reached();
-#else
-    if (snan_bit_is_one(status)) {
-# ifdef TARGET_HPPA
-        a &= ~0x0008000000000000ULL;
-        a |=  0x0004000000000000ULL;
-        return a;
-# else
-        return float64_default_nan(status);
-# endif
-    } else {
-        return a | LIT64(0x0008000000000000);
-    }
-#endif
-}
-
-
 /*----------------------------------------------------------------------------
 | Returns the result of converting the double-precision floating-point NaN
 | `a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
@@ -886,16 +819,10 @@ int floatx80_is_signaling_nan(floatx80 a, float_status *status)
 
 floatx80 floatx80_silence_nan(floatx80 a, float_status *status)
 {
-#ifdef NO_SIGNALING_NANS
-    g_assert_not_reached();
-#else
-    if (snan_bit_is_one(status)) {
-        return floatx80_default_nan(status);
-    } else {
-        a.low |= LIT64(0xC000000000000000);
-        return a;
-    }
-#endif
+    /* None of the targets that have snan_bit_is_one use floatx80.  */
+    assert(!snan_bit_is_one(status));
+    a.low |= LIT64(0xC000000000000000);
+    return a;
 }
 
 /*----------------------------------------------------------------------------
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index b5842f7b1c..40b039ee5b 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -2128,6 +2128,25 @@ float128 float128_default_nan(float_status *status)
     return r;
 }
 
+/*----------------------------------------------------------------------------
+| Returns a quiet NaN from a signalling NaN for the floating point value `a'.
+*----------------------------------------------------------------------------*/
+
+float16 float16_silence_nan(float16 a, float_status *status)
+{
+    return float16_pack_raw(parts_silence_nan(float16_unpack_raw(a), status));
+}
+
+float32 float32_silence_nan(float32 a, float_status *status)
+{
+    return float32_pack_raw(parts_silence_nan(float32_unpack_raw(a), status));
+}
+
+float64 float64_silence_nan(float64 a, float_status *status)
+{
+    return float64_pack_raw(parts_silence_nan(float64_unpack_raw(a), status));
+}
+
 /*----------------------------------------------------------------------------
 | Takes a 64-bit fixed-point value `absZ' with binary point between bits 6
 | and 7, and returns the properly rounded 32-bit integer corresponding to the
-- 
2.17.0

Re: [Qemu-devel] [PATCH v5 28/28] fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> Isolate the target-specific choice to 3 functions instead of 6.
>
> The code in floatx80_default_nan tried to be over-general.  There are
> only two targets that support this format: x86 and m68k.  Thus there
> is no point in inventing a mechanism for snan_bit_is_one.
>
> Move routines that no longer have ifdefs out of softfloat-specialize.h.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  fpu/softfloat-specialize.h | 81 ++------------------------------------
>  fpu/softfloat.c            | 19 +++++++++
>  2 files changed, 23 insertions(+), 77 deletions(-)
>
> diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h
> index ec4fb6ba8b..16c0bcb6fa 100644
> --- a/fpu/softfloat-specialize.h
> +++ b/fpu/softfloat-specialize.h
> @@ -278,24 +278,6 @@ int float16_is_signaling_nan(float16 a_, float_status *status)
>  #endif
>  }
>
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN from a signalling NaN for the half-precision
> -| floating point value `a'.
> -*----------------------------------------------------------------------------*/
> -
> -float16 float16_silence_nan(float16 a, float_status *status)
> -{
> -#ifdef NO_SIGNALING_NANS
> -    g_assert_not_reached();
> -#else
> -    if (snan_bit_is_one(status)) {
> -        return float16_default_nan(status);
> -    } else {
> -        return a | (1 << 9);
> -    }
> -#endif
> -}
> -
>  /*----------------------------------------------------------------------------
>  | Returns 1 if the single-precision floating-point value `a' is a quiet
>  | NaN; otherwise returns 0.
> @@ -334,30 +316,6 @@ int float32_is_signaling_nan(float32 a_, float_status *status)
>  #endif
>  }
>
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN from a signalling NaN for the single-precision
> -| floating point value `a'.
> -*----------------------------------------------------------------------------*/
> -
> -float32 float32_silence_nan(float32 a, float_status *status)
> -{
> -#ifdef NO_SIGNALING_NANS
> -    g_assert_not_reached();
> -#else
> -    if (snan_bit_is_one(status)) {
> -# ifdef TARGET_HPPA
> -        a &= ~0x00400000;
> -        a |=  0x00200000;
> -        return a;
> -# else
> -        return float32_default_nan(status);
> -# endif
> -    } else {
> -        return a | (1 << 22);
> -    }
> -#endif
> -}
> -
>  /*----------------------------------------------------------------------------
>  | Returns the result of converting the single-precision floating-point NaN
>  | `a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
> @@ -706,31 +664,6 @@ int float64_is_signaling_nan(float64 a_, float_status *status)
>  #endif
>  }
>
> -/*----------------------------------------------------------------------------
> -| Returns a quiet NaN from a signalling NaN for the double-precision
> -| floating point value `a'.
> -*----------------------------------------------------------------------------*/
> -
> -float64 float64_silence_nan(float64 a, float_status *status)
> -{
> -#ifdef NO_SIGNALING_NANS
> -    g_assert_not_reached();
> -#else
> -    if (snan_bit_is_one(status)) {
> -# ifdef TARGET_HPPA
> -        a &= ~0x0008000000000000ULL;
> -        a |=  0x0004000000000000ULL;
> -        return a;
> -# else
> -        return float64_default_nan(status);
> -# endif
> -    } else {
> -        return a | LIT64(0x0008000000000000);
> -    }
> -#endif
> -}
> -
> -
>  /*----------------------------------------------------------------------------
>  | Returns the result of converting the double-precision floating-point NaN
>  | `a' to the canonical NaN format.  If `a' is a signaling NaN, the invalid
> @@ -886,16 +819,10 @@ int floatx80_is_signaling_nan(floatx80 a, float_status *status)
>
>  floatx80 floatx80_silence_nan(floatx80 a, float_status *status)
>  {
> -#ifdef NO_SIGNALING_NANS
> -    g_assert_not_reached();
> -#else
> -    if (snan_bit_is_one(status)) {
> -        return floatx80_default_nan(status);
> -    } else {
> -        a.low |= LIT64(0xC000000000000000);
> -        return a;
> -    }
> -#endif
> +    /* None of the targets that have snan_bit_is_one use floatx80.  */
> +    assert(!snan_bit_is_one(status));
> +    a.low |= LIT64(0xC000000000000000);
> +    return a;
>  }
>
>  /*----------------------------------------------------------------------------
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index b5842f7b1c..40b039ee5b 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -2128,6 +2128,25 @@ float128 float128_default_nan(float_status *status)
>      return r;
>  }
>
> +/*----------------------------------------------------------------------------
> +| Returns a quiet NaN from a signalling NaN for the floating point value `a'.
> +*----------------------------------------------------------------------------*/
> +
> +float16 float16_silence_nan(float16 a, float_status *status)
> +{
> +    return float16_pack_raw(parts_silence_nan(float16_unpack_raw(a), status));
> +}
> +
> +float32 float32_silence_nan(float32 a, float_status *status)
> +{
> +    return float32_pack_raw(parts_silence_nan(float32_unpack_raw(a), status));
> +}
> +
> +float64 float64_silence_nan(float64 a, float_status *status)
> +{
> +    return float64_pack_raw(parts_silence_nan(float64_unpack_raw(a), status));
> +}
> +

Not that I'm objecting to the rationalisation but did you look at the
code generated now we unpack NaNs? I guess NaN behaviour isn't the
critical path for performance anyway....

Anyway:

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>


>  /*----------------------------------------------------------------------------
>  | Takes a 64-bit fixed-point value `absZ' with binary point between bits 6
>  | and 7, and returns the properly rounded 32-bit integer corresponding to the


--
Alex Bennée

Re: [Qemu-devel] [PATCH v5 28/28] fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan

[Richard Henderson]

On 05/15/2018 06:45 AM, Alex Bennée wrote:
>> +float64 float64_silence_nan(float64 a, float_status *status)
>> +{
>> +    return float64_pack_raw(parts_silence_nan(float64_unpack_raw(a), status));
>> +}
>> +
> 
> Not that I'm objecting to the rationalisation but did you look at the
> code generated now we unpack NaNs? I guess NaN behaviour isn't the
> critical path for performance anyway....

Yes, I looked.  It's about 5 instructions instead of 1.
But as you say, it's nowhere near critical path.

Ug.  I've also just realized that the shift isn't correct though...


r~

Re: [Qemu-devel] [PATCH v5 28/28] fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan

[Richard Henderson]

On 05/15/2018 08:41 AM, Richard Henderson wrote:
> On 05/15/2018 06:45 AM, Alex Bennée wrote:
>>> +float64 float64_silence_nan(float64 a, float_status *status)
>>> +{
>>> +    return float64_pack_raw(parts_silence_nan(float64_unpack_raw(a), status));
>>> +}
>>> +
>>
>> Not that I'm objecting to the rationalisation but did you look at the
>> code generated now we unpack NaNs? I guess NaN behaviour isn't the
>> critical path for performance anyway....
> 
> Yes, I looked.  It's about 5 instructions instead of 1.
> But as you say, it's nowhere near critical path.
> 
> Ug.  I've also just realized that the shift isn't correct though...

Having fixed that and re-checked... the compiler is weird.

The float32 version optimizes to 1 insn, as we would hope.  The float16 version
optimizes to 5 insns, extracting and re-inserting the sign bit.  The float64
version optimizes to 10 insns, extracting and re-inserting the exponent as well.

Very odd.


r~

Re: [Qemu-devel] [PATCH v5 00/28] softfloat patch roundup

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> This is my SNaN patch set, Alex's float-float refactor, and a couple
> of other random outstanding fpu patches.  This has been reordered so
> as to be bisectable, since the float-float refactor requires the snan
> work to avoid breakage.
>
> This was built on top of pm215/target-arm.next to make it easier to
> test.
> The whole tree is
>
>   git://github.com/rth7680/qemu.git fpu-roundup

Whole series:

Tested-by: Alex Bennée <alex.bennee@linaro.org>

Also I ran Emilio's benchmarks:

  https://i.imgur.com/Lww65Nx.png

Looks OK.

>
> Changes since last:
>   - Use v5 instead of v3, since Alex was up to v4 with float-float.
>   - Incorporate feedback from pm215.
>   - Include float128 fix from Petr.
>   - Include further cleanups for default_nan and silence_nan.
>     The first of these was inspired by a comment from pm215;
>     the rest follow as logical extensions.
>
>
> r~
>
>
> Alex Bennée (4):
>   target/arm: convert conversion helpers to fpst/ahp_flag
>   target/arm: squash FZ16 behaviour for conversions
>   fpu/softfloat: Partial support for ARM Alternative half-precision
>   fpu/softfloat: re-factor float to float conversions
>
> Petr Tesarik (1):
>   fpu/softfloat: Fix conversion from uint64 to float128
>
> Richard Henderson (23):
>   fpu/softfloat: Merge NO_SIGNALING_NANS definitions
>   fpu/softfloat: Split floatXX_silence_nan from floatXX_maybe_silence_nan
>   fpu/softfloat: Move softfloat-specialize.h below FloatParts definition
>   fpu/softfloat: Canonicalize NaN fraction
>   fpu/softfloat: Introduce parts_is_snan_frac
>   fpu/softfloat: Replace float_class_dnan with parts_default_nan
>   fpu/softfloat: Replace float_class_msnan with parts_silence_nan
>   target/arm: Use floatX_silence_nan when we have already checked for SNaN
>   target/arm: Remove floatX_maybe_silence_nan from conversions
>   target/hppa: Remove floatX_maybe_silence_nan from conversions
>   target/m68k: Use floatX_silence_nan when we have already checked for SNaN
>   target/mips: Remove floatX_maybe_silence_nan from conversions
>   target/riscv: Remove floatX_maybe_silence_nan from conversions
>   target/s390x: Remove floatX_maybe_silence_nan from conversions
>   fpu/softfloat: Use float*_silence_nan in propagateFloat*NaN
>   fpu/softfloat: Remove floatX_maybe_silence_nan
>   fpu/softfloat: Specialize on snan_bit_is_one
>   fpu/softfloat: Make is_nan et al available to softfloat-specialize.h
>   fpu/softfloat: Pass FloatClass to pickNaN
>   fpu/softfloat: Pass FloatClass to pickNaNMulAdd
>   fpu/softfloat: Define floatN_default_nan in terms of parts_default_nan
>   fpu/softfloat: Clean up parts_default_nan
>   fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan
>
>  fpu/softfloat-specialize.h    | 691 ++++++++++++++-------------------
>  include/fpu/softfloat-types.h |   1 +
>  include/fpu/softfloat.h       |  18 +-
>  target/arm/helper.h           |  10 +-
>  target/arm/translate.h        |  12 +
>  fpu/softfloat.c               | 708 ++++++++++++----------------------
>  target/arm/helper-a64.c       |   7 +-
>  target/arm/helper.c           | 114 +++---
>  target/arm/translate-a64.c    |  38 +-
>  target/arm/translate.c        |  74 +++-
>  target/hppa/cpu.c             |   1 -
>  target/hppa/op_helper.c       |   2 -
>  target/m68k/softfloat.c       |   3 +-
>  target/mips/msa_helper.c      |   4 -
>  target/mips/op_helper.c       |   2 -
>  target/ppc/fpu_helper.c       |   1 -
>  target/riscv/fpu_helper.c     |   6 +-
>  target/s390x/fpu_helper.c     |  12 +-
>  target/sh4/cpu.c              |   1 -
>  target/unicore32/cpu.c        |   2 -
>  20 files changed, 701 insertions(+), 1006 deletions(-)


--
Alex Bennée

Re: [Qemu-devel] [PATCH v5 00/28] softfloat patch roundup

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> This is my SNaN patch set, Alex's float-float refactor, and a couple
> of other random outstanding fpu patches.  This has been reordered so
> as to be bisectable, since the float-float refactor requires the snan
> work to avoid breakage.
>
> This was built on top of pm215/target-arm.next to make it easier to test.
> The whole tree is
>
>   git://github.com/rth7680/qemu.git fpu-roundup

I've finished my review pass, looking good.

>
> Changes since last:
>   - Use v5 instead of v3, since Alex was up to v4 with float-float.
>   - Incorporate feedback from pm215.
>   - Include float128 fix from Petr.
>   - Include further cleanups for default_nan and silence_nan.
>     The first of these was inspired by a comment from pm215;
>     the rest follow as logical extensions.
>
>
> r~
>
>
> Alex Bennée (4):
>   target/arm: convert conversion helpers to fpst/ahp_flag
>   target/arm: squash FZ16 behaviour for conversions
>   fpu/softfloat: Partial support for ARM Alternative half-precision
>   fpu/softfloat: re-factor float to float conversions
>
> Petr Tesarik (1):
>   fpu/softfloat: Fix conversion from uint64 to float128
>
> Richard Henderson (23):
>   fpu/softfloat: Merge NO_SIGNALING_NANS definitions
>   fpu/softfloat: Split floatXX_silence_nan from floatXX_maybe_silence_nan
>   fpu/softfloat: Move softfloat-specialize.h below FloatParts definition
>   fpu/softfloat: Canonicalize NaN fraction
>   fpu/softfloat: Introduce parts_is_snan_frac
>   fpu/softfloat: Replace float_class_dnan with parts_default_nan
>   fpu/softfloat: Replace float_class_msnan with parts_silence_nan
>   target/arm: Use floatX_silence_nan when we have already checked for SNaN
>   target/arm: Remove floatX_maybe_silence_nan from conversions
>   target/hppa: Remove floatX_maybe_silence_nan from conversions
>   target/m68k: Use floatX_silence_nan when we have already checked for SNaN
>   target/mips: Remove floatX_maybe_silence_nan from conversions
>   target/riscv: Remove floatX_maybe_silence_nan from conversions
>   target/s390x: Remove floatX_maybe_silence_nan from conversions
>   fpu/softfloat: Use float*_silence_nan in propagateFloat*NaN
>   fpu/softfloat: Remove floatX_maybe_silence_nan
>   fpu/softfloat: Specialize on snan_bit_is_one
>   fpu/softfloat: Make is_nan et al available to softfloat-specialize.h
>   fpu/softfloat: Pass FloatClass to pickNaN
>   fpu/softfloat: Pass FloatClass to pickNaNMulAdd
>   fpu/softfloat: Define floatN_default_nan in terms of parts_default_nan
>   fpu/softfloat: Clean up parts_default_nan
>   fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan
>
>  fpu/softfloat-specialize.h    | 691 ++++++++++++++-------------------
>  include/fpu/softfloat-types.h |   1 +
>  include/fpu/softfloat.h       |  18 +-
>  target/arm/helper.h           |  10 +-
>  target/arm/translate.h        |  12 +
>  fpu/softfloat.c               | 708 ++++++++++++----------------------
>  target/arm/helper-a64.c       |   7 +-
>  target/arm/helper.c           | 114 +++---
>  target/arm/translate-a64.c    |  38 +-
>  target/arm/translate.c        |  74 +++-
>  target/hppa/cpu.c             |   1 -
>  target/hppa/op_helper.c       |   2 -
>  target/m68k/softfloat.c       |   3 +-
>  target/mips/msa_helper.c      |   4 -
>  target/mips/op_helper.c       |   2 -
>  target/ppc/fpu_helper.c       |   1 -
>  target/riscv/fpu_helper.c     |   6 +-
>  target/s390x/fpu_helper.c     |  12 +-
>  target/sh4/cpu.c              |   1 -
>  target/unicore32/cpu.c        |   2 -
>  20 files changed, 701 insertions(+), 1006 deletions(-)


--
Alex Bennée