[Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Richard Henderson]

Having fixed the UL vs ULL fiasco in patch 17.


r~


The following changes since commit 385057cbec9b4a0eb6150330c572e875ed714965:

  Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2015-05-15' into staging (2015-05-15 17:51:20 +0100)

are available in the git repository at:

  git://github.com/rth7680/qemu.git tags/pull-axp-20150521

for you to fetch changes up to 32ad48abd74a997220b841e4e913edeb267aa362:

  target-alpha: Add vector implementation for CMPBGE (2015-05-21 10:34:18 -0700)

----------------------------------------------------------------
Rewrite fp exceptions

----------------------------------------------------------------
Richard Henderson (18):
      target-alpha: Move VAX helpers to a new file
      target-alpha: Rename floating-point subroutines
      target-alpha: Forget installed round mode after MT_FPCR
      target-alpha: Set PC correctly for floating-point exceptions
      target-alpha: Tidy FPCR representation
      target-alpha: Set fpcr_exc_status even for disabled exceptions
      target-alpha: Set EXC_M_SWC for exceptions from /S insns
      target-alpha: Raise IOV from CVTTQ
      target-alpha: Fix cvttq vs large integers
      target-alpha: Fix cvttq vs inf
      target-alpha: Fix integer overflow checking insns
      target-alpha: Implement WH64EN
      target-alpha: Disallow literal operand to 1C.30 to 1C.37
      target-alpha: Raise EXC_M_INV properly for fp inputs
      target-alpha: Suppress underflow from CVTTQ if DNZ
      target-alpha: Raise IOV from CVTQL
      target-alpha: Rewrite helper_zapnot
      target-alpha: Add vector implementation for CMPBGE

 target-alpha/Makefile.objs |   2 +-
 target-alpha/cpu.h         |  95 ++++----
 target-alpha/fpu_helper.c  | 530 +++++++++++----------------------------------
 target-alpha/helper.c      | 132 ++---------
 target-alpha/helper.h      |  14 +-
 target-alpha/int_helper.c  | 126 +++++------
 target-alpha/mem_helper.c  |   9 +-
 target-alpha/translate.c   | 265 ++++++++++++-----------
 target-alpha/vax_helper.c  | 353 ++++++++++++++++++++++++++++++
 9 files changed, 752 insertions(+), 774 deletions(-)
 create mode 100644 target-alpha/vax_helper.c

[Qemu-devel] [PULL v2 17/18] target-alpha: Rewrite helper_zapnot

[Richard Henderson]

This form produces significantly smaller code on x86_64.

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 target-alpha/int_helper.c | 30 ++++++++++++------------------
 1 file changed, 12 insertions(+), 18 deletions(-)

diff --git a/target-alpha/int_helper.c b/target-alpha/int_helper.c
index 8e4537f..74f38cb 100644
--- a/target-alpha/int_helper.c
+++ b/target-alpha/int_helper.c
@@ -37,31 +37,25 @@ uint64_t helper_cttz(uint64_t arg)
     return ctz64(arg);
 }
 
-static inline uint64_t byte_zap(uint64_t op, uint8_t mskb)
+uint64_t helper_zapnot(uint64_t val, uint64_t mskb)
 {
     uint64_t mask;
 
-    mask = 0;
-    mask |= ((mskb >> 0) & 1) * 0x00000000000000FFULL;
-    mask |= ((mskb >> 1) & 1) * 0x000000000000FF00ULL;
-    mask |= ((mskb >> 2) & 1) * 0x0000000000FF0000ULL;
-    mask |= ((mskb >> 3) & 1) * 0x00000000FF000000ULL;
-    mask |= ((mskb >> 4) & 1) * 0x000000FF00000000ULL;
-    mask |= ((mskb >> 5) & 1) * 0x0000FF0000000000ULL;
-    mask |= ((mskb >> 6) & 1) * 0x00FF000000000000ULL;
-    mask |= ((mskb >> 7) & 1) * 0xFF00000000000000ULL;
-
-    return op & ~mask;
-}
+    mask  = -(mskb & 0x01) & 0x00000000000000ffull;
+    mask |= -(mskb & 0x02) & 0x000000000000ff00ull;
+    mask |= -(mskb & 0x04) & 0x0000000000ff0000ull;
+    mask |= -(mskb & 0x08) & 0x00000000ff000000ull;
+    mask |= -(mskb & 0x10) & 0x000000ff00000000ull;
+    mask |= -(mskb & 0x20) & 0x0000ff0000000000ull;
+    mask |= -(mskb & 0x40) & 0x00ff000000000000ull;
+    mask |= -(mskb & 0x80) & 0xff00000000000000ull;
 
-uint64_t helper_zap(uint64_t val, uint64_t mask)
-{
-    return byte_zap(val, mask);
+    return val & mask;
 }
 
-uint64_t helper_zapnot(uint64_t val, uint64_t mask)
+uint64_t helper_zap(uint64_t val, uint64_t mask)
 {
-    return byte_zap(val, ~mask);
+    return helper_zapnot(val, ~mask);
 }
 
 uint64_t helper_cmpbge(uint64_t op1, uint64_t op2)
-- 
2.1.0

Re: [Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Peter Maydell]

On 21 May 2015 at 18:39, Richard Henderson <rth@twiddle.net> wrote:
> Having fixed the UL vs ULL fiasco in patch 17.

...you seem to have changed the implementation too?
The previous pull used sextract64().

-- PMM

Re: [Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Richard Henderson]

On 05/21/2015 01:08 PM, Peter Maydell wrote:
> On 21 May 2015 at 18:39, Richard Henderson <rth@twiddle.net> wrote:
>> Having fixed the UL vs ULL fiasco in patch 17.
> 
> ...you seem to have changed the implementation too?
> The previous pull used sextract64().

Yes.  This is the same number of insns on x86_64, but doesn't overcommit the
shifter.  I should have mentioned.


r~

Re: [Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Peter Maydell]

On 21 May 2015 at 18:39, Richard Henderson <rth@twiddle.net> wrote:
> Having fixed the UL vs ULL fiasco in patch 17.
>
>
> r~
>
>
> The following changes since commit 385057cbec9b4a0eb6150330c572e875ed714965:
>
>   Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2015-05-15' into staging (2015-05-15 17:51:20 +0100)
>
> are available in the git repository at:
>
>   git://github.com/rth7680/qemu.git tags/pull-axp-20150521
>
> for you to fetch changes up to 32ad48abd74a997220b841e4e913edeb267aa362:
>
>   target-alpha: Add vector implementation for CMPBGE (2015-05-21 10:34:18 -0700)

Applied, thanks.

-- PMM

Re: [Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Peter Maydell]

On 22 May 2015 at 11:00, Peter Maydell <peter.maydell@linaro.org> wrote:
> On 21 May 2015 at 18:39, Richard Henderson <rth@twiddle.net> wrote:
>> for you to fetch changes up to 32ad48abd74a997220b841e4e913edeb267aa362:
>>
>>   target-alpha: Add vector implementation for CMPBGE (2015-05-21 10:34:18 -0700)
>
> Applied, thanks.

Except I failed to notice a test failure from one of my builds
before pushing:

target-alpha/int_helper.c: In function 'helper_cmpbge':
target-alpha/int_helper.c:77: error: invalid operands to binary >=
target-alpha/int_helper.c:79: error: subscripted value is neither
array nor pointer

That's in the Centos5 build, which is
gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-55)

Any suggestions?

I do note that this build target is basically obsolete anyway, because
we're about to move to requiring glib 2.22, at which point Centos5 is
officially Too Old. So maybe we just say "yeah, you need a newer gcc" ?

-- PMM

Re: [Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Peter Maydell]

On 22 May 2015 at 11:22, Peter Maydell <peter.maydell@linaro.org> wrote:
> On 22 May 2015 at 11:00, Peter Maydell <peter.maydell@linaro.org> wrote:
>> On 21 May 2015 at 18:39, Richard Henderson <rth@twiddle.net> wrote:
>>> for you to fetch changes up to 32ad48abd74a997220b841e4e913edeb267aa362:
>>>
>>>   target-alpha: Add vector implementation for CMPBGE (2015-05-21 10:34:18 -0700)
>>
>> Applied, thanks.
>
> Except I failed to notice a test failure from one of my builds
> before pushing:
>
> target-alpha/int_helper.c: In function 'helper_cmpbge':
> target-alpha/int_helper.c:77: error: invalid operands to binary >=
> target-alpha/int_helper.c:79: error: subscripted value is neither
> array nor pointer
>
> That's in the Centos5 build, which is
> gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-55)
>
> Any suggestions?
>
> I do note that this build target is basically obsolete anyway, because
> we're about to move to requiring glib 2.22, at which point Centos5 is
> officially Too Old. So maybe we just say "yeah, you need a newer gcc" ?

Hmm. This fails on some of the travis builds too, with slightly
different warnings:

gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3
target-alpha/int_helper.c:77:24: error: invalid operands to binary >=
(have '__vector(16) unsigned char' and '__vector(16) unsigned char')

I think the best thing to do for the moment is to revert commit
32ad48abd74a9 so we can get trunk building again, so I'm going to
do that.

thanks
-- PMM

Re: [Qemu-devel] [PULL v2 00/18] target-alpha fpu improvements

[Richard Henderson]

On 05/22/2015 04:32 AM, Peter Maydell wrote:
> On 22 May 2015 at 11:22, Peter Maydell <peter.maydell@linaro.org> wrote:
>> On 22 May 2015 at 11:00, Peter Maydell <peter.maydell@linaro.org> wrote:
>>> On 21 May 2015 at 18:39, Richard Henderson <rth@twiddle.net> wrote:
>>>> for you to fetch changes up to 32ad48abd74a997220b841e4e913edeb267aa362:
>>>>
>>>>   target-alpha: Add vector implementation for CMPBGE (2015-05-21 10:34:18 -0700)
>>>
>>> Applied, thanks.
>>
>> Except I failed to notice a test failure from one of my builds
>> before pushing:
>>
>> target-alpha/int_helper.c: In function 'helper_cmpbge':
>> target-alpha/int_helper.c:77: error: invalid operands to binary >=
>> target-alpha/int_helper.c:79: error: subscripted value is neither
>> array nor pointer
>>
>> That's in the Centos5 build, which is
>> gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-55)
>>
>> Any suggestions?
>>
>> I do note that this build target is basically obsolete anyway, because
>> we're about to move to requiring glib 2.22, at which point Centos5 is
>> officially Too Old. So maybe we just say "yeah, you need a newer gcc" ?
> 
> Hmm. This fails on some of the travis builds too, with slightly
> different warnings:
> 
> gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3
> target-alpha/int_helper.c:77:24: error: invalid operands to binary >=
> (have '__vector(16) unsigned char' and '__vector(16) unsigned char')
> 
> I think the best thing to do for the moment is to revert commit
> 32ad48abd74a9 so we can get trunk building again, so I'm going to
> do that.

Wow, I would have thought 4.6 would have been new enough.
Sorry for the headache.  Yes, a revert is probably best
until I can figure out what versions do what.


r~