Julian Seward [Sat, 26 Jan 2013 11:47:55 +0000 (11:47 +0000)]
Infrastructure cleanup: change type of the condition field of
IRExpr_Mux0X from Ity_I8 to Ity_I1. This makes more sense, makes it
consistent with condition fields in IRStmt_Dirty and IRStmt_Exit, and
avoids some pointless 1Uto8 casting of the condition, in many cases.
Fixes for s390 are from Florian.
Also, make a small extension to ir_opt.c, that allows the constant
folder to look backwards through arbitrary expressions even in flat
IR. This makes it possible to do arbitrary tree folding in ir_opt,
which is where it belongs. Use this to implement the folding rule
CmpNE32(1Uto32(b), 0) ==> b.
Julian Seward [Fri, 25 Jan 2013 09:46:43 +0000 (09:46 +0000)]
Annotate ARMNImm_to_Imm64 with fallthrough markers following
verification against the table in host_arm_defs.h, "Neon Immediate
operand". A particularly nasty piece of code.
Carl Love [Tue, 22 Jan 2013 20:25:31 +0000 (20:25 +0000)]
Fix implementation of the DFP integer operands.
The implementation of integer operands doesn't really match the documentation
for the Iop. Take for example Iop_ExtractExpD64. It is documented as
D64 -> I64 but the implementation of the UNARY is defined as
UNARY(Ity_D64, Ity_D64). The result is an integer that is stored in an integer
format in a floating point register. On the IBM s390 however, the architecture
stores the integer value in a general purpose register (GPR) not a floating
point register. This issue exists with the implementation of 11 Iops where the
PPC implementation has either a source or destination whose value is an integer
but the value is stored in a floating point register in an integer format. After
reviewing the PPC implementation with the s390 developer, it was agreed the
cleanest way to fix this is to change the PPC implementation. The BINOP will be
changed to be consistent with the Iop description. This means the PPC
instruction implementation of the PPC instruction in guest_ppc_toIR.c will need
to reinterpret integer source operands as integers which will move the value
from a floating point register to an integer register before calling binop().
The underlying PPC implementation of the unop() for the specific Iop will also
need to change to move the value from the integer register back to the floating
point register so the native instruction can be issued with the integer value
in a floating point register. It was decided that making the changed in PPC,
rather then having the s390 reinterpret integers as DFP and then move the value
back to an integer register, was preferable as it makes the implementation of
the unop(), binops(), triop() consistent with the definition of the Iop.
This patch also includes the needed changes for the vbit tester. The Iop
definitions in memcheck/tests/vbit-test/util.c had to be updated to be consitent
with the changes in the Iops as documented below. Also, the function mkLazy3()
in memcheck/mc_translate.c had to be updated to handle the I32 x I8 x I64 -> I64
and I32 x I8 x I128 -> I128 cases.
The specific list of changes are as follows:
Iop name in pub/libvex_ir.h
documented type
type of UNARY/BINARY/TERNARY in priv/ir_defs.c
-------------------------------------------------------
Carl Love [Mon, 21 Jan 2013 18:12:31 +0000 (18:12 +0000)]
The 32-bit DFP value is stored in a 64-bit register in
ppc. The D32 to D64 and D64 to D32 definitions for the
Iop type was specified in VEX/priv/ir_defs.c, function
typeOfPrimop() as:
case Iop_D32toD64:
UNARY(Ity_64, Ity_D64);
case Iop_D64toD32:
BINARY(ity_RMode, Ity_D64, Ity_D64);
since the values resided in a 64-bit register. As part of the s390 DFP support
the definitions were changed to:
case Iop_D32toD64:
UNARY(Ity_32, Ity_D64);
case Iop_D64toD32:
BINARY(ity_RMode, Ity_D64, Ity_D32);
to reflect what they really should be. However, this broke the ppc
implementation. Valgrind would fail and report a mismatch on the types as the
ppc code was using a D64 instead of a D32.
This patch adds support for fetching and storing the Dfp32 operand as a 32-bit
value. The support includes adding the functions iselDfp32Expr() and
iselDfp32Expr_wrk() and additional code to support the DFP32 bit iops.
Petar Jovanovic [Sun, 20 Jan 2013 18:16:45 +0000 (18:16 +0000)]
mips: fix for mips-disassembler when branch is at block_size-2 position
Check if the last instruction in the block is a branch or jump instruction
should happen only if the disassembler was not already stopped.
Incorrect conditional led to a boundary case in which jumps/branches were not
executed when placed on "max_insns - 2" position in the block.
none/tests/mips32/block_size test will be added to Valgrind to describe the case
and check for regressions in future.
Florian Krohm [Sun, 20 Jan 2013 03:51:04 +0000 (03:51 +0000)]
Improve the tree builder in IR optimisation. Allow load expressions to be
moved past Put/I statements and dirty helpers, when it is safe to do so.
It is safe, when the statement does not require exact memory exceptions.
New functions stmt_modifies_guest_state and dirty_helper_puts have been
added to determine the side effect on the guest state.
This optimisation enables the use of memory-to-memory insns on
architectures that have those.
Note that this macro behaves slightly differently for some types
from the gcc __alignof__ and from the equivalent (new) standardised
alignof.
vg_alignof macro is needed for the "perm_malloc" callers (next commit)
to determine the alignment of small blocks, but might be useful
for other purposes => placed in libvex_basictypes.h, close to offsetof.
Julian Seward [Wed, 16 Jan 2013 22:11:13 +0000 (22:11 +0000)]
mips32 equivalent to r2636: fix up the mips32 back end to be in sync
with infrastructural changes w.r.t. conditional dirty helpers that
return values. Does not actually handle such cases since the mips32
front end does not generate them.
Julian Seward [Wed, 16 Jan 2013 21:10:01 +0000 (21:10 +0000)]
Fix up the s390 back end to be in sync with infrastructural changes
w.r.t. conditional dirty helpers that return values. Does not
actually handle such cases since the s390 front end does not generate
them. Further ahead, it would be more general to redo this by
incorporating a RetLoc as part of the helper_call struct. This change
is OK for now, though.
Julian Seward [Wed, 16 Jan 2013 14:56:06 +0000 (14:56 +0000)]
ppc32/64 equivalents to r2636: fix up the ppc back end to be in sync
with infrastructural changes w.r.t. conditional dirty helpers that
return values. Does not actually handle such cases since the ppc
front end does not generate them.
Julian Seward [Wed, 16 Jan 2013 09:29:37 +0000 (09:29 +0000)]
x86 equivalent to r2636: fix up the x86 back end to be in sync with
infrastructural changes w.r.t. conditional dirty helpers that return
values. Does not actually handle such cases since the x86 front end
does not generate them.
Julian Seward [Tue, 15 Jan 2013 22:30:39 +0000 (22:30 +0000)]
Fix up the amd64 back end to be in sync with infrastructural changes
w.r.t. conditional dirty helpers that return values. Does not
actually handle such cases since the amd64 front end does not generat
them.
Florian Krohm [Sun, 13 Jan 2013 02:29:05 +0000 (02:29 +0000)]
s390: Support insns to convert between DFP values and signed/unsigned
integers. Patch by Maran Pakkirisamy (maranp@linux.vnet.ibm.com).
Part of fixing BZ 307113.
Florian Krohm [Sat, 12 Jan 2013 22:02:07 +0000 (22:02 +0000)]
Add 12 IROps for converting betwen DFP values and signed/unsigned integers.
Patch by Maran Pakkirisamy (maranp@linux.vnet.ibm.com).
Part of fixing BZ 307113.
Julian Seward [Tue, 8 Jan 2013 14:09:04 +0000 (14:09 +0000)]
Get rid of selectable default (return) values in conditional dirty
calls, as introduced in r2594. It is overkill -- unnecessary
complexity. Instead have a pre-assumed default bit pattern of 0101010
(0x55..) to be returned in such cases.
Carl Love [Thu, 3 Jan 2013 23:34:18 +0000 (23:34 +0000)]
The call to set the rounding mode for DFP iops: Iop_AddD128, Iop_SubD128,
Iop_MulD128, Iop_DivD128, and Iop_D128toI64 is wrong. The call being used is
set_FPU_rounding_mode(). This call is used to set the two rounding mode bits
for the Floating point instructions. The call set_FPU_DFP_rounding_mode()
should have been used to set the three rounding mode bits for the DFP
instructions.
This patch changes the call to the correct function to set the DFP
rounding mode bits.
Florian Krohm [Sun, 30 Dec 2012 18:17:18 +0000 (18:17 +0000)]
Improve handling of dirty helper calls when building trees in ado_treebuild_BB.
This function took an overly conservative approach and always assumed
that calling a dirty helper would modify both guest state and memory. This
patch introduces two new functions dirty_helper_stores and dirty_helper_puts,
to determine the actual side effets of a helper call. Using these functions
increases precision and allows the tree builder to move a GET past a dirty
helper call.
Florian Krohm [Thu, 27 Dec 2012 20:14:03 +0000 (20:14 +0000)]
s390: Support the "test data class/group" and "extract significance"
insns. Patch by Maran Pakkirisamy (maranp@linux.vnet.ibm.com).
This is part of fixing BZ 307113.
Florian Krohm [Thu, 27 Dec 2012 00:59:43 +0000 (00:59 +0000)]
s390: Do not waste a register when assigning a constant to a memory
location. If available, use MVHI and friends. If those are not available,
load the constant value into register r0 and store that. r0 is not visible
to register allocation and therefore using it does not increase register
pressure.
Remove S390_INSN_MZERO and replace it with S390_INSN_MIMM. Assigning zero
is just a special case..
Saves between 0.9% and 2.4% of insns as measured with the perf regression
bucket.
Florian Krohm [Wed, 26 Dec 2012 17:47:19 +0000 (17:47 +0000)]
s390x: Support "compare biased exponent" insns CEDTR, CEXTR.
To do that properly, two new IROps are needed: Iop_CmpExpD64 and
Iop_CmpExpD128. It might seem that extracting the exponents using
Iop_ExtractExpD64/D128 and comparing the values could be used here.
But that only works for finite DFP values. Hence, the new IROps.
Patch by Maran Pakkirisamy (maranp@linux.vnet.ibm.com).
This is part of fixing BZ 307113.
Florian Krohm [Mon, 24 Dec 2012 00:14:31 +0000 (00:14 +0000)]
s390: New file s390_defs.h to contain definitions that are
neither guest nor host specific, but just s390 specific. These
definitions formerly resided in host_s390_defs.h
Florian Krohm [Sun, 23 Dec 2012 01:09:16 +0000 (01:09 +0000)]
s390: Simplify the optimisation for PUTs to tracked registers.
There is no need to treat the guest IA register specially. The first
assignment in a SB to any tracked register will always assign the full value
and not do an incremental update. Therefore, maintaining
env->first_IA_assignment is redundant. Interestingly, the old code
forgot to mark the tracking state as valid when first_IA_assignment was
true, resulting in an inefficiency (in the order of 0.8% of insns).
Also, when a tracked register already contains the value that it is
being assigned, there is not need to
(a) check whether that register requires precise memory exceptions and
(b) if so, assign the new value.
By definition, the register is up-to-date and reassigning the new value
which is also the old value is meaningless. No idea, what I was thinking
at the time.
Florian Krohm [Sat, 22 Dec 2012 15:01:04 +0000 (15:01 +0000)]
s390: Reduce the size of an s390_insn from 48 bytes to 32 bytes by
- adding one level of indirection for CDAS and DFP binops
- rearranging and tweaking the variant for helper calls
- using seperate insns kinds for signed and unsigned multiplication
and division.
Add an assert to make sure the size stays at 32 bytes.
Florian Krohm [Sat, 22 Dec 2012 02:28:25 +0000 (02:28 +0000)]
s390: Generalise S390_INSN_GADD which only worked on the guest
state to S390_INSN_MADD which works for any memory location
addressable with base reg + 20-bit displacement.
Florian Krohm [Fri, 21 Dec 2012 21:05:17 +0000 (21:05 +0000)]
s390: Distinguish between conversion to/from IRCmpFxxResult and
IRCmpDxxResult, even though the encodings are currently the same.
Rename convert_s390_fpcc_to_vex to convert_s390_to_vex_bfpcc.
Add convert_s390_to_vex_dfpcc and convert_vex_dfpcc_to_s390.
Florian Krohm [Fri, 21 Dec 2012 17:32:12 +0000 (17:32 +0000)]
Add support for these DFP insns:
AXTRA, CDTR, CXTR, DXTRA, LDETR, LXDTR, LDXTR, LEDTR, LTXTR, MXTRA, SXTRA
This is part of fixing BZ #307113.
Patch by Maran Pakkirisamy (maranp@linux.vnet.ibm.com) with some minor
mods.
Florian Krohm [Thu, 20 Dec 2012 16:58:52 +0000 (16:58 +0000)]
s390: Fix IR generation for DDTR and MDTR. These insns do not modify
the condition code.
Spotted and fixed by Maran Pakkirisamy (maranp@linux.vnet.ibm.com).
Julian Seward [Mon, 17 Dec 2012 21:54:10 +0000 (21:54 +0000)]
IR: add the ability to specify the call-didn't-happen ("default")
return value for conditional dirty helper calls returning values. The
default value can be either all-zeroes or all-ones.
Florian Krohm [Sun, 16 Dec 2012 22:49:05 +0000 (22:49 +0000)]
Fix an assert in s390 disassembly. Eliminate a few magic array
width constants and use S390_MAX_MNEMONIC_LEN instead.
New function "mnemonic" to construct a mnemonic string padded with
blanks for alignment.
Florian Krohm [Tue, 11 Dec 2012 04:09:43 +0000 (04:09 +0000)]
Generalise S390_INSN_GZERO which only worked on the guest
state to S390_INSN_MZERO which works for any memory location
addressable with base reg + 12-bit displacement.
Florian Krohm [Tue, 4 Dec 2012 04:45:32 +0000 (04:45 +0000)]
In the past, the implementation of STFLE returned the facilities of the host
machine. This was not consistent in the following sense: Suppose the host
has a facility F installed and this facility implies the availability of an
insn X. Suppose further, that insn X is not supported in valgrind.
An application progrm that tests the availability of insn X by checking
for its associated facility F will fail under valgrind when using X because
valgrind will SIGILL. Not so good.
This patch changes the STFLE behaviour to adjust the facilities of the
virtual machine according to what the set of insns that is actually
supported. It's an approximation, because for some facilities we only
support a subset of the insns enabled by that facility.
Julian Seward [Sun, 25 Nov 2012 15:26:48 +0000 (15:26 +0000)]
Initial front changes for ARM, to generate direct IR for at least some
conditional loads and stores. Very incomplete -- most load-store
cases still use the old scheme.
projects / thirdparty / valgrind.git / log
