Patchwork [V4,04/22] softfloat: Fix float64_to_uint32

login
register
mail settings
Submitter Tom Musta
Date Dec. 18, 2013, 8:19 p.m.
Message ID <1387397961-4894-5-git-send-email-tommusta@gmail.com>
Download mbox | patch
Permalink /patch/303088/
State New
Headers show

Comments

Tom Musta - Dec. 18, 2013, 8:19 p.m.
The float64_to_uint32 has several flaws:

 - for numbers between 2**32 and 2**64, the inexact exception flag
   may get incorrectly set.  In this case, only the invalid flag
   should be set.

       test pattern: 425F81378DC0CD1F / 0x1.f81378dc0cd1fp+38

 - for numbers between 2**63 and 2**64, incorrect results may
   be produced:

       test pattern: 43EAAF73F1F0B8BD / 0x1.aaf73f1f0b8bdp+63

This patch re-implements float64_to_uint32 to re-use the
float64_to_uint64 routine (instead of float64_to_int64).  For the
saturation case, the inexact bit is explicitly cleared before raising
the invalid flag, provided that it was not previously set.

V4: Fixed handling of stickiness of the inexact bit per comments from
Peter Maydell.

This contribution can be licensed under either the softfloat-2a or -2b
license.

Signed-off-by: Tom Musta <tommusta@gmail.com>
---
 fpu/softfloat.c |   15 +++++++--------
 1 files changed, 7 insertions(+), 8 deletions(-)
Peter Maydell - Dec. 19, 2013, 9:48 p.m.
On 18 December 2013 20:19, Tom Musta <tommusta@gmail.com> wrote:
> The float64_to_uint32 has several flaws:
>
>  - for numbers between 2**32 and 2**64, the inexact exception flag
>    may get incorrectly set.  In this case, only the invalid flag
>    should be set.
>
>        test pattern: 425F81378DC0CD1F / 0x1.f81378dc0cd1fp+38
>
>  - for numbers between 2**63 and 2**64, incorrect results may
>    be produced:
>
>        test pattern: 43EAAF73F1F0B8BD / 0x1.aaf73f1f0b8bdp+63
>
> This patch re-implements float64_to_uint32 to re-use the
> float64_to_uint64 routine (instead of float64_to_int64).  For the
> saturation case, the inexact bit is explicitly cleared before raising
> the invalid flag, provided that it was not previously set.
>
> V4: Fixed handling of stickiness of the inexact bit per comments from
> Peter Maydell.
>
> This contribution can be licensed under either the softfloat-2a or -2b
> license.
>
> Signed-off-by: Tom Musta <tommusta@gmail.com>

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

(though you might tidy up the commit message as per other
email).

I have a patch which fixes the remaining float-to-int conversion
functions (ie the ones you didn't deal with in this series) to not
raise Inexact when they raise Invalid for out of range, and one
which fixes a bug in scalbn. I'll send those out shortly.

thanks
-- PMM

Patch

diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index 1b614ae..6110e28 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -6578,19 +6578,18 @@  uint_fast16_t float32_to_uint16_round_to_zero(float32 a STATUS_PARAM)
 
 uint32 float64_to_uint32( float64 a STATUS_PARAM )
 {
-    int64_t v;
+    uint64_t v;
     uint32 res;
+    int old_exc_flags = get_float_exception_flags(status);
 
-    v = float64_to_int64(a STATUS_VAR);
-    if (v < 0) {
-        res = 0;
-        float_raise( float_flag_invalid STATUS_VAR);
-    } else if (v > 0xffffffff) {
+    v = float64_to_uint64(a STATUS_VAR);
+    if (v > 0xffffffff) {
         res = 0xffffffff;
-        float_raise( float_flag_invalid STATUS_VAR);
     } else {
-        res = v;
+        return v;
     }
+    set_float_exception_flags(old_exc_flags, status);
+    float_raise(float_flag_invalid STATUS_VAR);
     return res;
 }