Exception handling in fmadd/fmsub/fnmadd/fnmsub isnt correct as the
order of checking could give wrong settings in FPSCR.
For example, (x * y) + z, if x = infinity, y = zero and z = snan.
After the execution of instruction VXNAN and VXIMZ both should be set. For
this correct the ordering in the float64_maddsub_update_excp() as follows:
* If x, y or z is an SNaN, vxsnan_flag is set to 1.
* If x is a Zero and y, is an Infinity or x is an Infinity and y is
an Zero, vximz_flag is set to 1.
* If the product of x and y is an Infinity and z is an Infinity of
the opposite sign, vxisi_flag is set to 1.
Moreover, all vector multiply-add/substract and vector scalar multiply-add/sub
had the bug. VXISI should be set only when (∞ - ∞), where as the instruction was
doing it wrong, was just checking ((a == ∞ OR b == ∞) && (c == ∞)). There are
two
issues here:
1. infinity can be +ve or -ve, i.e. (∞ + (-∞)), should result in setting VXISI
2. Need to take care of the operation (add or sub). (∞ + ∞) should not set VXISI
Patch:
01: Fixes the order of checking and makes them independent as per
the ISA
02: Introduces the macro to be used by Vector Scalar and Vector Multiply-Add
03: Make Vector Scalar and Vector Multiply Add use the macro for updating
exception flags
Changelog:
v0:
* Use ALL CAPS for macro arguments
* Correct the argument type in the function
Nikunj A Dadhania (3):
target/ppc: fmadd check for excp independently
target/ppc: fmadd: add macro for updating flags
target/ppc: use helper for excp handling
target/ppc/fpu_helper.c | 77 ++++++++++++++++++++-----------------------------
1 file changed, 32 insertions(+), 45 deletions(-)
--
2.7.4
