softfloat: Improve denormal handling

[PATCH 00/11] softfloat: Improve denormal handling

Richard Henderson posted 11 patches 4 years, 8 months ago

Download series mbox

Test checkpatch passed

Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/20210527041405.391567-1-richard.henderson@linaro.org

Maintainers: Aurelien Jarno <aurelien@aurel32.net>, "Philippe Mathieu-Daudé" <f4bug@amsat.org>, "Alex Bennée" <alex.bennee@linaro.org>, Peter Maydell <peter.maydell@linaro.org>, Jiaxun Yang <jiaxun.yang@flygoat.com>, Eduardo Habkost <ehabkost@redhat.com>, Aleksandar Rikalo <aleksandar.rikalo@syrmia.com>, Yoshinori Sato <ysato@users.sourceforge.jp>, Bastian Koppelmann <kbastian@mail.uni-paderborn.de>, Paolo Bonzini <pbonzini@redhat.com>, Richard Henderson <richard.henderson@linaro.org>

include/fpu/softfloat-types.h |  18 +-
fpu/softfloat.c               |  88 ++++------
target/arm/sve_helper.c       |   6 +-
target/arm/vfp_helper.c       |  12 +-
target/i386/tcg/fpu_helper.c  |  18 +-
target/mips/tcg/msa_helper.c  | 298 +++++++++++++++-------------------
target/rx/op_helper.c         |  19 ++-
target/tricore/fpu_helper.c   |   6 +-
fpu/softfloat-parts.c.inc     |  13 +-
9 files changed, 212 insertions(+), 266 deletions(-)

Expand all Fold all

[PATCH 00/11] softfloat: Improve denormal handling

Posted by Richard Henderson 4 years, 8 months ago

This attempts to fix the x86 denormal-exception flag, which is the
inverse of the existing float_flag_input_denormal flag.  I have not
created a new test case for this yet, fwiw.

While auditing all uses of float_flag_*_denormal, I found some errors
in target/rx and target/mips.

This is based on my current softfloat conversion for floatx80:
Based-on: <20210525150706.294968-1-richard.henderson@linaro.org>

The complete tree may be found at
https://gitlab.com/rth7680/qemu/-/commits/fpu-test-5


r~


Richard Henderson (11):
  softfloat: Rename float_flag_input_denormal to
    float_flag_iflush_denormal
  softfloat: Rename float_flag_output_denormal to
    float_flag_oflush_denormal
  softfloat: Introduce float_flag_inorm_denormal
  softfloat: Introduce float_flag_result_denormal
  target/i386: Use float_flag_inorm_denormal
  target/rx: Handle the FPSW.DN bit in helper_set_fpsw
  target/rx: Use FloatRoundMode in helper_set_fpsw
  target/rx: Fix setting of FPSW.CE
  target/mips: Drop inline markers from msa_helper.c
  target/mips: Do not check MSACSR_FS_MASK in update_msacsr
  target/mips: Drop denormal operand to update_msacsr

 include/fpu/softfloat-types.h |  18 +-
 fpu/softfloat.c               |  88 ++++------
 target/arm/sve_helper.c       |   6 +-
 target/arm/vfp_helper.c       |  12 +-
 target/i386/tcg/fpu_helper.c  |  18 +-
 target/mips/tcg/msa_helper.c  | 298 +++++++++++++++-------------------
 target/rx/op_helper.c         |  19 ++-
 target/tricore/fpu_helper.c   |   6 +-
 fpu/softfloat-parts.c.inc     |  13 +-
 9 files changed, 212 insertions(+), 266 deletions(-)

-- 
2.25.1

Re: [PATCH 00/11] softfloat: Improve denormal handling

Posted by Alex Bennée 4 years, 8 months ago

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

> This attempts to fix the x86 denormal-exception flag, which is the
> inverse of the existing float_flag_input_denormal flag.  I have not
> created a new test case for this yet, fwiw.
>
> While auditing all uses of float_flag_*_denormal, I found some errors
> in target/rx and target/mips.

Softfloat code looks good to me, I'm leaving the target specific stuff
to the arch maintainers as I have a hard enough time with just one
architectures FP vagaries ;-)

-- 
Alex Bennée