This patch modifies SoftFloat library so that it can be configured in
run-time in relation to the meaning of signaling NaN bit, while, at the
same time, strictly preserving its behavior on all existing platforms.
Background:
In floating-point calculations, there is a need for denoting undefined or
unrepresentable values. This is achieved by defining certain floating-point
numerical values to be NaNs (which stands for "not a number"). For additional
reasons, virtually all modern floating-point unit implementations use two
kinds of NaNs: quiet and signaling. The binary representations of these two
kinds of NaNs, as a rule, differ only in one bit (that bit is, traditionally,
the first bit of mantissa).
Up to 2008, standards for floating-point did not specify all details about
binary representation of NaNs. More specifically, the meaning of the bit
that is used for distinguishing between signaling and quiet NaNs was not
strictly prescribed. (IEEE 754-2008 was the first floating-point standard
that defined that meaning clearly, see [1], p. 35) As a result, different
platforms took different approaches, and that presented considerable
challenge for multi-platform emulators like QEMU.
Mips platform represents the most complex case among QEMU-supported
platforms regarding signaling NaN bit. Up to the Release 6 of Mips
architecture, "1" in signaling NaN bit denoted signaling NaN, which is
opposite to IEEE 754-2008 standard. From Release 6 on, Mips architecture
adopted IEEE standard prescription, and "0" denotes signaling NaN. On top of
that, Mips architecture for SIMD (also known as MSA, or vector instructions)
also specifies signaling bit in accordance to IEEE standard. MSA unit can be
implemented with both pre-Release 6 and Release 6 main processor units.
QEMU uses SoftFloat library to implement various floating-point-related
instructions on all platforms. The current QEMU implementation allows for
defining meaning of signaling NaN bit during build time, and is implemented
via preprocessor macro called SNAN_BIT_IS_ONE.
On the other hand, the change in this patch enables SoftFloat library to be
configured in run-time. This configuration is meant to occur during CPU
initialization, at the moment when it is definitely known what desired
behavior for particular CPU (or any additional FPUs) is.
The change is implemented so that it is consistent with existing
implementation of similar cases. This means that structure float_status is
used for passing the information about desired signaling NaN bit on each
invocation of SoftFloat functions. The additional field in float_status is
called snan_bit_is_one, which supersedes macro SNAN_BIT_IS_ONE.
IMPORTANT:
This change is not meant to create any change in emulator behavior or
functionality on any platform. It just provides the means for SoftFloat
library to be used in a more flexible way - in other words, it will just
prepare SoftFloat library for usage related to Mips platform and its
specifics regarding signaling bit meaning, which is done in some of
subsequent patches from this series.
Further break down of changes:
1) Added field snan_bit_is_one to the structure float_status, and
correspondent setter function set_snan_bit_is_one().
2) Constants <float16|float32|float64|floatx80|float128>_default_nan
(used both internally and externally) converted to functions
<float16|float32|float64|floatx80|float128>_default_nan(float_status*).
This is necessary since they are dependent on signaling bit meaning.
At the same time, for the sake of code cleanup and simplicity, constants
<floatx80|float128>_default_nan_<low|high> (used only internally within
SoftFloat library) are removed, as not needed.
3) Added a float_status* argument to SoftFloat library functions
XXX_is_quiet_nan(XXX a_), XXX_is_signaling_nan(XXX a_),
XXX_maybe_silence_nan(XXX a_). This argument must be present in
order to enable correct invocation of new version of functions
XXX_default_nan(). (XXX is <float16|float32|float64|floatx80|float128>
here)
4) Updated code for all platforms to reflect changes in SoftFloat library.
This change is twofolds: it includes modifications of SoftFloat library
functions invocations, and an addition of invocation of function
set_snan_bit_is_one() during CPU initialization, with arguments that
are appropriate for each particular platform. It was established that
all platforms zero their main CPU data structures, so snan_bit_is_one(0)
in appropriate places is not added, as it is not needed.
[1] "IEEE Standard for Floating-Point Arithmetic",
IEEE Computer Society, August 29, 2008.
Signed-off-by: Thomas Schwinge <thomas@codesourcery.com>
Signed-off-by: Maciej W. Rozycki <macro@codesourcery.com>
Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Tested-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Leon Alrae <leon.alrae@imgtec.com>
Tested-by: Leon Alrae <leon.alrae@imgtec.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
[leon.alrae@imgtec.com:
* cherry-picked 2 chunks from patch #2 to fix compilation warnings]
Signed-off-by: Leon Alrae <leon.alrae@imgtec.com>
With specification at hand from the reference manual from Freescale
http://cache.nxp.com/files/32bit/doc/ref_manual/SPEPEM.pdf , I have found a fix
to efscmp* instructions handling in QEMU.
efscmp* instructions in QEMU set crD (Condition Register nibble) values as
(0b0100 << 2) = 0b10000 (consider the HELPER_SINGLE_SPE_CMP macro which left
shifts the value returned by efscmp* handler by 2 bits). A value of 0b10000 is
not correct according the to the reference manual.
The reference manual expects efscmp* instructions to return a value of 0bx1xx.
Please find attached a patch which disables left shifting in
HELPER_SINGLE_SPE_CMP macro. This macro is used by efscmp* and efstst*
instructions only. efstst* instruction handlers, in turn, call efscmp* handlers
too.
*Explanation:*
Traditionally, each crD (condition register nibble) consist of 4 bits, which is
set by comparisons as follows:
crD = W X Y Z
where
W = Less than
X = Greater than
Y = Equal to
However, efscmp* instructions being a special case return a binary result.
(efscmpeq will set the crD = 0bx1xx iff when op1 == op2 and 0bx0xx otherwise;
i.e. there is no notion of different crD values based on Less than, Greater
than and Equal to).
This effectively means that crD will store a "Greater than" comparison result
iff efscmp* instruction comparison is TRUE. Compiler exploits this feature by
checking for "Branch if Less than or Equal to" (ble instruction) OR "Branch if
Greater than" (bgt instruction) for Branch if FALSE OR Branch if TRUE
respectively after an efscmp* instruction. This can be seen in a assembly code
snippet below:
27 if (__real__ x != 3.0f || __imag__ x != 4.0f)
10000498: lwz r10,8(r31)
1000049c: lis r9,16448
100004a0: efscmpeq cr7,r10,r9
100004a4: ble- cr7,0x100004b8 <bar+60> //jump to abort() call
100004a8: lwz r10,12(r31)
100004ac: lis r9,16512
100004b0: efscmpeq cr7,r10,r9
100004b4: bgt- cr7,0x100004bc <bar+64> //skip abort() call
28 abort ();
100004b8: bl 0x10000808 <abort>
Signed-off-by: Talha Imran <talha_imran@mentor.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.
This commit was created with scripts/clean-includes.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-6-git-send-email-peter.maydell@linaro.org
Currently in TCG mode, updating floating exception
summary bit (FPSCR_FX) in fpscr also updates
the upper 32bits of fpscr with all 1s.
Modify the bit shift operation statement to use
1ULL instead.
Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The set_fprf argument to the helper_compute_fprf helper function
is no longer necessary -- the helper is only invoked when FPSCR[FPRF]
is going to be set.
Eliminate the unnecessary argument from the function signature and
its corresponding implementation. Change the return value of the
helper to "void". Update the name of the local variable "ret" to
"fprf", which now makes more sense.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
The Power ISA square root instructions (fsqrt[s], frsqrte[s]) must
set the FPSCR[VXSQRT] flag when operating on a negative value.
However, NaNs have no sign and therefore this flag should not
be set when operating on one.
Change the order of the checks in the helper code. Move the
SNaN-to-QNaN macro to the top of the file so that it can be
re-used.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch remove limitation for fc[tf]id[*] on 32 bits targets and
add a new insn flag for signed integer 64 conversion PPC2_FP_CVT_S64
Signed-off-by: Pierre Mallard <mallard.pierre@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Remove the code that reduce the result to float32 as the frsqrte
instruction is defined to return a double-precision estimate of
the reciprocal square root.
Although reducing the fractional part is harmless (as the estimation
must have at least 12 bits of precision according to the old PEM),
reducing the exponent range is not correct.
Signed-off-by: Tristan Gingold <gingold@adacore.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Rather than include helper.h with N values of GEN_HELPER, include a
secondary file that sets up the macros to include helper.h. This
minimizes the files that must be rebuilt when changing the macros
for file N.
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <rth@twiddle.net>
This patch corrects the VSX integer to floating point conversion instructions
by using the endian correct accessors. The auxiliary "j" index used by the
existing macros is now obsolete and is removed. The JOFFSET preprocessor
macro is also obsolete and removed.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch corrects the VSX floating point to integer conversion
instructions by using the endian correct accessors. The auxiliary
"j" index used by the existing macros is now obsolete and is removed.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This change corrects the VSX double precision to single precision and
single precision to double precisions conversion routines. The endian
correct accessors are now used. The auxiliary "j" index is no longer
necessary and is eliminated.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This change fixes the VSX scalar compare instructions. The existing usage of "x.f64[0]"
is changed to "x.VsrD(0)".
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
A common pattern in the VSX helper code macros is the use of "x.fld[i]" where
"x" is a VSR and "fld" is an argument to a macro ("f64" or "f32" is passed).
This is not always correct on LE hosts.
This change addresses all instances of this pattern to be "x.fld" where "fld" is:
- "VsrD(0)" for scalar instructions accessing 64-bit numbers
- "VsrD(i)" for vector instructions accessing 64-bit numbers
- "VsrW(i)" for vector instructions accessing 32-bit numbers
Note that there are no instances of this pattern where a scalar instruction
accesses a 32-bit number.
Note also that it would be correct to use "VsrD(i)" for scalar instructions since
the loop index is only ever "0". I have choosen to use "VsrD(0)" instead ... it
seems a little clearer.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This change properly orders the doublewords of the VSRs 0-31. Because these
registers are constructed from separate doublewords, they must be inverted
on Little Endian hosts. The inversion is performed both when the VSR is read
and when it is written.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This change defines accessors for VSR doubleword and word fields that
are correct from a host Endian perspective. This allows code to
use the Power ISA indexing numbers in code.
For example, the xscvdpsxws instruction has a target VSR that looks
like this:
0 32 64 127
+-----------+--------+-----------+-----------+
| undefined | SW | undefined | undefined |
+-----------+--------+-----------+-----------+
VSX helper code will use VsrW(1) to access this field.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
The various VSX Convert to Integer instructions should truncate the
floating point number to an integer value, which is equivalent to
a round-to-zero rounding mode. The existing VSX floating point to
integer conversion helpers are erroneously using the rounding mode set
int the PowerPC Floating Point Status and Control Register (FPSCR).
This change corrects this defect by using the appropriate
float*_to_*_round_to_zero() routines fro the softfloat library.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Tested-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch fixes 64 bit constants that were erroneously declared as "ul" instead of
"ull". The preferred form "ULL" is used.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the Floating Point Test for Square Root instruction
which was introduced in Power ISA 2.06.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the Floating Point Test for Divide instruction which
was introduced in Power ISA 2.06B.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
The fri* series of instructions was introduced prior to ISA 2.06 and
is supported on Power7 and Power8 hardware. However, the instruction
is still considered illegal in the P7 and P8 QEMU emulation models.
This patch enables these instructions for the P7 and P8 machines.
Also, the existing helper is modified to correctly handle some of
the boundary cases (NaNs and the inexact flag).
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the fcfids, fcfidu and fcfidus instructions which
were introduced in Power ISA 2.06B. A common macro is provided to
eliminate repetitious code, and the existing fcfid instruction is
refactored to use this macro.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the four floating point to integer conversion instructions
introduced by Power ISA V2.06:
- Floating Convert to Integer Word Unsigned (fctiwu)
- Floating Convert to Integer Word Unsigned with Round Toward
Zero (fctiwuz)
- Floating Convert to Integer Doubleword Unsigned (fctidu)
- Floating Convert to Integer Doubleword Unsigned with Round
Toward Zero (fctiduz)
A common macro is developed to eliminate repetitious code. Existing instructions
are also refactoried to use this macro (fctiw, fctiwz, fctid, fctidz).
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the non-signalling scalar conversion instructions:
- VSX Scalar Convert Single Precision to Double Precision
Non-Signalling (xscvspdpn)
- VSX Scalar Convert Double Precision to Single Precision
Non-Signalling (xscvdpspn)
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Round to Single Precision (xsrsp)
instruction.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Convert Unsigned Integer Doubleword
to Floating Point Format and Round to Single Precision (xscvuxdsp)
and VSX Scalar Convert Signed Integer Douglbeword to Floating Point
Format and Round to Single Precision (xscvsxdsp) instructions.
The existing integer to floating point conversion macro (VSX_CVT_INT_TO_FP)
is modified to support the rounding of the intermediate floating point
result to single precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the Single Precision VSX Scalar Fused Multiply-Add
instructions: xsmaddasp, xsmaddmsp, xssubasp, xssubmsp, xsnmaddasp,
xsnmaddmsp, xsnmsubasp, xsnmsubmsp.
The existing VSX_MADD() macro is modified to support rounding of the
intermediate double precision result to single precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Reciprocal Square Root Estimate
Single Precision (xsrsqrtesp) instruction.
The existing VSX_RSQRTE() macro is modified to support rounding
of the intermediate double-precision result to single precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Square Root Single Precision (xssqrtsp)
instruction.
The existing VSX_SQRT() macro is modified to support rounding of the
intermediate double-precision result to single-precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Reciprocal Estimate Single Precision
(xsresp) instruction.
The existing VSX_RE macro is modified to support rounding of the
intermediate double precision result to single precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Divide Single Precision (xsdivsp)
instruction.
The existing VSX_DIV macro is modified to support rounding of the
intermediate double precision result to single precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Multiply Single-Precision (xsmulsp)
instruction.
The existing VSX_MUL macro is modified to support rounding of the
intermediate result to single precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX Scalar Add Single-Precision (xsaddsp) and
VSX Scalar Subtract Single-Precision (xssubsp) instructions.
The existing VSX_ADD_SUB macro is modified to support the rounding
of the (intermediate) result to single-precision.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX instructions that convert between floating
point formats: xscvdpsp, xscvspdp, xvcvdpsp, xvcvspdp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point compare vector instructions:
- xvcmpeqdp[.], xvcmpgedp[.], xvcmpgtdp[.]
- xvcmpeqsp[.], xvcmpgesp[.], xvcmpgtsp[.]
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point maximum and minimum
instructions:
- xsmaxdp, xvmaxdp, xvmaxsp
- xsmindp, xvmindp, xvminsp
Because of the Power ISA definitions of maximum and minimum
on various boundary cases, the standard softfloat comparison
routines (e.g. float64_lt) do not work as well as one might
think. Therefore specific routines for comparing 64 and 32
bit floating point numbers are implemented in the PowerPC
helper code.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX scalar floating point compare ordered
and unordered instructions.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point test for software square
root instructions defined by V2.06 of the PowerPC ISA: xstsqrtdp,
xvtsqrtdp, xvtsqrtsp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point test for software divide
instructions defined by V2.06 of the PowerPC ISA: xstdivdp, xvtdivdp,
and xvtdivsp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point reciprocal square root
estimate instructions defined by V2.06 of the PowerPC ISA: xsrsqrtedp,
xvrsqrtedp, xvrsqrtesp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point square root instructions
defined by V2.06 of the PowerPC ISA: xssqrtdp, xvsqrtdp, xvsqrtsp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point reciprocal estimate instructions
defined by V2.06 of the PowerPC ISA: xsredp, xvredp, xvresp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point divide instructions defined
by V2.06 of the PowerPC ISA: xsdivdp, xvdivdp, xvdivsp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the VSX floating point multiply instructions defined
by V2.06 of the PowerPC ISA: xsmuldp, xvmuldp, xvmulsp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds the floating point addition and subtraction
instructions defined by V2.06 of the PowerPC ISA: xssubdp,
xvsubdp and xvsubsp.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds general support that will be used by the VSX helper
routines:
- a union describing the various VSR subfields.
- access routines to get and set VSRs
- VSX decoders
- a general routine to generate a handler that invokes a VSX
helper.
Signed-off-by: Tom Musta <tommusta@gmail.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
