[Qemu-devel] [PATCH v3] softfloat: Add round-to-odd rounding mode

Bharata B Rao posted 1 patch 7 years, 1 month ago
Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/1486360515-22989-1-git-send-email-bharata@linux.vnet.ibm.com
Test checkpatch passed
Test docker passed
Test s390x passed
fpu/softfloat.c         | 21 ++++++++++++++++++++-
include/fpu/softfloat.h |  2 ++
2 files changed, 22 insertions(+), 1 deletion(-)
[Qemu-devel] [PATCH v3] softfloat: Add round-to-odd rounding mode
Posted by Bharata B Rao 7 years, 1 month ago
Power ISA 3.0 introduces a few quadruple precision floating point
instructions that support round-to-odd rounding mode. The
round-to-odd mode is explained as under:

Let Z be the intermediate arithmetic result or the operand of a convert
operation. If Z can be represented exactly in the target format, the
result is Z. Otherwise the result is either Z1 or Z2 whichever is odd.
Here Z1 and Z2 are the next larger and smaller numbers representable
in the target format respectively.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
---
Tested the corner cases for round-to-odd and the default round-to-nearest-even
by comparing the results of QEMU with a known good implementation of
PowerPC ISA 3.0 for the following instructions: xsaddqp[o], xsdivqp[o],
xsmulqp[o] and xscvqpdp[o].

Changes in v3:

- Limited the recalculation of roundIncrement in 64bit rounding only
  to round-to-odd case and hence dropped the separate patch which
  did this for all rounding modes earlier. (Peter Maydell)
- Resorted to QEMU codying style in softfloat.c

v2: https://www.mail-archive.com/qemu-devel@nongnu.org/msg425741.html

 fpu/softfloat.c         | 21 ++++++++++++++++++++-
 include/fpu/softfloat.h |  2 ++
 2 files changed, 22 insertions(+), 1 deletion(-)

diff --git a/fpu/softfloat.c b/fpu/softfloat.c
index c295f31..5ccba76 100644
--- a/fpu/softfloat.c
+++ b/fpu/softfloat.c
@@ -623,6 +623,9 @@ static float64 roundAndPackFloat64(flag zSign, int zExp, uint64_t zSig,
     case float_round_down:
         roundIncrement = zSign ? 0x3ff : 0;
         break;
+    case float_round_to_odd:
+        roundIncrement = (zSig & 0x400) ? 0 : 0x3ff;
+        break;
     default:
         abort();
     }
@@ -632,8 +635,10 @@ static float64 roundAndPackFloat64(flag zSign, int zExp, uint64_t zSig,
              || (    ( zExp == 0x7FD )
                   && ( (int64_t) ( zSig + roundIncrement ) < 0 ) )
            ) {
+            bool overflow_to_inf = roundingMode != float_round_to_odd &&
+                                   roundIncrement != 0;
             float_raise(float_flag_overflow | float_flag_inexact, status);
-            return packFloat64( zSign, 0x7FF, - ( roundIncrement == 0 ));
+            return packFloat64(zSign, 0x7FF, -(!overflow_to_inf));
         }
         if ( zExp < 0 ) {
             if (status->flush_to_zero) {
@@ -651,6 +656,13 @@ static float64 roundAndPackFloat64(flag zSign, int zExp, uint64_t zSig,
             if (isTiny && roundBits) {
                 float_raise(float_flag_underflow, status);
             }
+            if (roundingMode == float_round_to_odd) {
+                /*
+                 * For round-to-odd case, the roundIncrement depends on
+                 * zSig which just changed.
+                 */
+                roundIncrement = (zSig & 0x400) ? 0 : 0x3ff;
+            }
         }
     }
     if (roundBits) {
@@ -1149,6 +1161,9 @@ static float128 roundAndPackFloat128(flag zSign, int32_t zExp,
     case float_round_down:
         increment = zSign && zSig2;
         break;
+    case float_round_to_odd:
+        increment = !(zSig1 & 0x1) && zSig2;
+        break;
     default:
         abort();
     }
@@ -1168,6 +1183,7 @@ static float128 roundAndPackFloat128(flag zSign, int32_t zExp,
             if (    ( roundingMode == float_round_to_zero )
                  || ( zSign && ( roundingMode == float_round_up ) )
                  || ( ! zSign && ( roundingMode == float_round_down ) )
+                 || (roundingMode == float_round_to_odd)
                ) {
                 return
                     packFloat128(
@@ -1215,6 +1231,9 @@ static float128 roundAndPackFloat128(flag zSign, int32_t zExp,
             case float_round_down:
                 increment = zSign && zSig2;
                 break;
+            case float_round_to_odd:
+                increment = !(zSig1 & 0x1) && zSig2;
+                break;
             default:
                 abort();
             }
diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h
index 842ec6b..8a39028 100644
--- a/include/fpu/softfloat.h
+++ b/include/fpu/softfloat.h
@@ -180,6 +180,8 @@ enum {
     float_round_up           = 2,
     float_round_to_zero      = 3,
     float_round_ties_away    = 4,
+    /* Not an IEEE rounding mode: round to the closest odd mantissa value */
+    float_round_to_odd       = 5,
 };
 
 /*----------------------------------------------------------------------------
-- 
2.7.4


Re: [Qemu-devel] [PATCH v3] softfloat: Add round-to-odd rounding mode
Posted by Richard Henderson 7 years, 1 month ago
On 02/05/2017 09:55 PM, Bharata B Rao wrote:
> Power ISA 3.0 introduces a few quadruple precision floating point
> instructions that support round-to-odd rounding mode. The
> round-to-odd mode is explained as under:
>
> Let Z be the intermediate arithmetic result or the operand of a convert
> operation. If Z can be represented exactly in the target format, the
> result is Z. Otherwise the result is either Z1 or Z2 whichever is odd.
> Here Z1 and Z2 are the next larger and smaller numbers representable
> in the target format respectively.
>
> Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> ---
> Tested the corner cases for round-to-odd and the default round-to-nearest-even
> by comparing the results of QEMU with a known good implementation of
> PowerPC ISA 3.0 for the following instructions: xsaddqp[o], xsdivqp[o],
> xsmulqp[o] and xscvqpdp[o].
>
> Changes in v3:
>
> - Limited the recalculation of roundIncrement in 64bit rounding only
>   to round-to-odd case and hence dropped the separate patch which
>   did this for all rounding modes earlier. (Peter Maydell)
> - Resorted to QEMU codying style in softfloat.c

Reviewed-by: Richard Henderson <rth@twiddle.net>


r~