[OpenVMS documentation]
[Site home] [Send comments] [Help with this site] [How to order documentation] [OpenVMS site] [Compaq site]
Updated: 11 December 1998

OpenVMS RTL Library (LIB$) Manual


Previous Contents Index

Some instructions can generate more than one fault if evaluation of one operand causes a fault that occurs before a later operand (which would also cause a fault). An example of this is the possibility that a floating-point divide instruction might report a divide-by-zero fault upon seeing a zero divisor before noticing that the dividend was a reserved operand or was inaccessible.

In these cases, operand-specific faults are signaled immediately by LIB$DECODE_FAULT in the expectation that another condition handler (or the same one) can repair the situation. This may reorder the sequence of exceptions as seen by a program. If the operand exception is corrected, the original exception reoccurs, and the proper action is taken.

If at all possible, try to determine if a resignal is necessary inside the condition handler that calls LIB$DECODE_FAULT, rather than inside your user action routine. The reason for this is that LIB$DECODE_FAULT removes all post-signal stack frames before calling your user action routine.

Your user action routine may fetch and store the operands, registers, and PSL as necessary for handling the exception. You should follow the VAX architecture rule of reading all input operands in left-to-right order, then writing all output operands in left-to-right order, to avoid inconsistent results with overlapping operands. This is especially necessary with register operands.

PSL may be modified in a manner consistent with the VAX architecture. If the T-bit in the PSL was set at the beginning of the instruction, LIB$DECODE_FAULT sets the TP bit. To initiate tracing, you must set only the T bit. To disable tracing, you must clear both the T and TP bits. See the VAX Architecture Reference Manual for more information.

If the first-part-done (FPD) bit in the PSL was set when the instruction faulted, LIB$DECODE_FAULT only advances the PC over the instruction; it does not reevaluate the operands, and it sets operand-count to zero. It is assumed that if FPD is set, the operands are in known locations (typically the registers).

For the CASEB, CASEW, and CASEL instructions, only the selector, base, and limit operands are represented in operand-count and read-operand-locations. The element of registers that corresponds to the PC, described in the following text as R15, points to the first of the word-length displacements. Your user action routine must modify R15 to reflect the location of the next instruction to execute.

The standard instruction definitions used by LIB$DECODE_FAULT specify the XFC instruction (which causes an SS$_OPCCUS fault) as having zero operands. You may redefine XFC if needed using the instruction-definitions argument to LIB$DECODE_FAULT.

If you do not want instruction execution to resume with the next sequential instruction, you must modify R15 appropriately. Your user action routine then returns to LIB$DECODE_FAULT, which restores the registers and PSL, and resumes instruction execution. See also the LIB$_RESTART condition value in the section called Condition Values Returned from the User Action Routine.

Note

Vector context is not saved or restored.

Exceptions Recognized by LIB$DECODE_FAULT

LIB$DECODE_FAULT recognizes the following VAX faults:

All other exceptions, including SS$_COMPAT and SS$_RADRMOD, cause LIB$DECODE_FAULT to return immediately with the return status SS$_RESIGNAL.

SS$_COMPAT is generated by compatibility-mode instructions. LIB$DECODE_FAULT does not handle compatibility-mode instructions.

SS$_RADRMOD is generated by a reserved addressing-mode fault. LIB$DECODE_FAULT assumes that all instructions follow VAX addressing-mode specifications.

Instruction Operand Definition Codes

Each instruction operand has an access type (read, write, ...) and a data type (byte, word, ...) associated with it. The operand definition codes used in both the instruction-definitions argument passed to LIB$DECODE_FAULT and in the operand-types argument passed to the user action routine encode the access and data types in a byte. The fields and values for operand access and data types are described using the symbols in Table lib-3. These symbols are defined in definition libraries supplied by Compaq as macro or module name $LIBDCFDEF.

Table lib-3 Symbols for Fields and Values for Operand Access and Data Types Using LIB$DECODE_FAULT
Symbol Description
LIB$V_DCFACC The field of the operand description code that describes the operand access type (bits 0--2).
LIB$S_DCFACC The size of the access type field (3 bits).
LIB$M_DCFACC The mask for the access type field. This is a 3-bit field that can contain any binary value from 000 through 111. The integer value of these bit settings defines the operand access type code for the LIB$M_DCFACC field. Currently, six codes are defined. These codes have symbolic names and are explained below. It is important to remember that LIB$M_DCFACC is not a bit mask. The values 0 through 6 do not refer to bits 0 through 6. They represent the binary values 001 through 110 as contained in the 3-bit field.

The operand access type codes defined for the LIB$M_DCFACC field are:
LIB$K_DCFACC_R = 1 Operand is read-only.
LIB$K_DCFACC_M = 2 Operand is to be modified.
LIB$K_DCFACC_W = 3 Operand is write-only.
LIB$K_DCFACC_A = 4 Operand is an address (must not be a register).
LIB$K_DCFACC_V = 5 Operand is the base of a bit field (same as address except that it may be a register).
LIB$K_DCFACC_B = 6 Operand is a branch address.

LIB$V_DCFTYP The field of the operand descriptor code that describes the operand data type (bits 3--7).
LIB$S_DCFTYP The size of the operand data type field (5 bits).
LIB$M_DCFTYP The mask for the operand data type field. This is a 5-bit field (bits 3--7) that can contain any binary value from 00000 through 11111. The integer value of these bit settings defines the operand access type code for the LIB$M_DCFACC field. Currently, nine codes are defined. These codes have symbolic names and are explained below. It is important to remember that LIB$M_DCFTYP is not a bit mask. The values 0 through 9 do not refer to bits 0 through 9. They represent the binary values 00001 through 01001 as contained in the 5-bit field. The operand access type codes defined for the LIB$V_DCFTYP field are:
LIB$K_DCFTYP_B = 1 Operand is a byte.
LIB$K_DCFTYP_W = 2 Operand is a word.
LIB$K_DCFTYP_L = 3 Operand is a longword.
LIB$K_DCFTYP_Q = 4 Operand is a quadword.
LIB$K_DCFTYP_O = 5 Operand is an octaword.
LIB$K_DCFTYP_F = 6 Operand is F_floating.
LIB$K_DCFTYP_D = 7 Operand is D_floating.
LIB$K_DCFTYP_G = 8 Operand is G_floating.
LIB$K_DCFTYP_H = 9 Operand is H_floating.

Symbols of the form LIB$K_DCFOPR_xy, where x is the access type and y is the data type, are also defined. These combine the notions of access and data type. For example, LIB$K_DCFOPR_MF has the following value:


50 (2+(6*8)) 

It denotes modify access of an F_floating item. For the branch access type, only the types BB, BW, and BL are defined; otherwise, all combinations are available.

Call Format for a User Action Routine

LIB$DECODE_FAULT calls the user action routine when it finds an exception to be handled. Your user action routine handles the exception in any manner that you specify and then returns to LIB$DECODE_FAULT.

action-routine opcode ,instr-PC ,PSL ,registers ,operand-count
,operand-types ,read-operand-locations
,write-operand-locations ,signal-arguments
,signal-procedure ,context
,unspecified-user-argument ,original-registers

opcode


OpenVMS usage: longword_unsigned
type: longword (unsigned)
access: read only
mechanism: by reference

Opcode of the instruction that caused the fault. The opcode argument is the address of a longword that contains this opcode. LIB$DECODE_FAULT supplies this opcode when it calls the user action routine.

For 2-byte opcodes, the escape code (for example, hex FD) is in the low-order byte. You must use this argument to examine the opcode instead of reading the bytes pointed to by instr-PC. This is because if a debugger breakpoint has been set on the instruction, only opcode contains the original instruction.

instr-PC


OpenVMS usage: longword_unsigned
type: longword (unsigned)
access: read only
mechanism: by reference

Value of the PC for the instruction that caused the fault. The instr-PC argument is the address of a longword that contains the PC value.

Note the difference between this value and the contents of the registers array element that corresponds to the PC. R15 of the registers array element contains the address of the byte after the instruction that caused the fault.

PSL


OpenVMS usage: longword_unsigned
type: longword (unsigned)
access: modify
mechanism: by reference

Processor status longword (PSL) at the time of the fault. The PSL argument is the address of a longword that contains this PSL. Your user action routine may modify this PSL within the restrictions of the VAX architecture.

registers


OpenVMS usage: vector_longword_unsigned
type: longword (unsigned)
access: modify
mechanism: by reference, array reference

Contents of registers R0 through R15 (PC) at the time of the fault but after operand addressing-mode processing. This includes any autoincrements or autodecrements. The registers argument is the address of this 16-longword array. Each longword of the registers array contains the contents of one register.

Your user action routine may modify these values. If it does, the new values will be reflected when instruction execution continues.

To modify vector registers, execute a vector instruction. Executing a vector instruction in the handler modifies the state of the vector processor. The state of the vector processor is not restored when the handler returns. This has the effect of altering the state when the execution continues.

R15 denotes the sixteenth longword in the registers array, which corresponds to the PC. R15 contains the address of the next byte after the current instruction. Unless this value is modified by your user action routine, instruction execution will resume at that address. An exception is for the CASEB, CASEW, and CASEL instructions; R15 contains the address of the first displacement word. For these instructions, your user action routine must modify R15 to point to the next instruction to execute.

Upon instruction completion, registers R0-R15 are restored from this array. However, if signal-procedure is used to cause a fault or if instruction restart is specified by returning LIB$_RESTART, original-registers is used instead.

operand-count


OpenVMS usage: longword_unsigned
type: longword (unsigned)
access: read only
mechanism: by reference

Number of operands in the instruction currently being decoded. The operand-count is the address of a longword that contains this number.

operand-types


OpenVMS usage: vector_longword_unsigned
type: longword (unsigned)
access: read only
mechanism: by reference, array reference

Array of longwords, one element for each operand, that contains the type codes for the associated operand. The operand-types argument is the address of this array.

The operand type codes are further defined in the section called Instruction Operand Definition Codes.

read-operand-locations


OpenVMS usage: vector_longword_unsigned
type: longword (unsigned)
access: read only
mechanism: by reference, array reference

Array of longwords, one element for each operand, that contains the addresses of the operands to be read. The read-operand-locations argument is the address of this array.

The address given in the array may not be the actual address of the operand if the operand is not a memory location. If the operand is a register, the address indicates a copy of the register values at the time of operand evaluation. If the operand access type is ADDRESS or FIELD and the operand is not a register, the address is the address of the item. If the operand access type is FIELD and the operand is a register, the address refers to the appropriate element in the registers array. If the operand access type is BRANCH, the address is the destination PC of the branch. For WRITE access operands, the address value is zero.

write-operand-locations


OpenVMS usage: vector_longword_unsigned
type: longword (unsigned)
access: read only
mechanism: by reference, array reference

Array of longwords, one element for each operand, that contains the addresses of operands that are to be written. The write-operand-locations argument is the address of this array. If the operand access type is not MODIFY, WRITE, ADDRESS, or FIELD, the pointer value is zero.

signal-arguments


OpenVMS usage: vector_longword_unsigned
type: longword (unsigned)
access: read only
mechanism: by reference, array reference

Signal arguments list of the original exception, as passed from OpenVMS to your condition handler and then to LIB$DECODE_FAULT. The signal-arguments argument is the address of an array of longwords that contains these signal arguments.

signal-procedure


OpenVMS usage: procedure
type: procedure value
access: call without stack unwinding
mechanism: by reference

Entry mask of a routine that your user action routine must call if it wants to report an exception for the instruction that faulted. The signal-procedure argument is the address of this entry mask.

For further information, see the section called Call Format for a Signal Routine.

context


OpenVMS usage: context
type: unspecified
access: read only
mechanism: by value

Context in which the exception occurs, including the register and PSL contents, to be used when calling the signal-procedure. The context argument contains the value of this context.

unspecified-user-argument


OpenVMS usage: user_arg
type: longword (unsigned)
access: read only
mechanism: by value

Optional argument passed to LIB$DECODE_FAULT. If the argument was not specified, the value zero is substituted. The unspecified-user-argument argument contains the value of this optional argument.

original-registers


OpenVMS usage: vector_longword_unsigned
type: longword (unsigned)
access: modify
mechanism: by reference, array reference

Array containing the values of registers R0 through R15 (PC) at the time of the fault, before operand processing. The original-registers argument is the address of this 16-longword array.

If the action routine specifies that the instruction should restart or that a fault should be generated, the registers are restored from original-registers. See also the description of registers above.

Condition Values Returned from the User Action Routine

The user action routine can return the following condition values to LIB$DECODE_FAULT:
Condition Value Description
SS$_CONTINUE If the user action routine returns a value of SS$_CONTINUE, instruction execution will continue as specified by the current contents of the registers element for the PC.
SS$_RESIGNAL If the user action routine returns SS$_RESIGNAL, the original exception is resignaled, with the only changes reflected being those specified by registers elements for R0 and R1 (which are stored in the mechanism arguments vector), PC, and PSL. All other registers are restored from original registers.
LIB$_RESTART If the user action routine returns LIB$_RESTART, the current instruction is restarted with registers restored from original-registers and a PSL from PSL. This feature is useful for writing trace handlers.

Call Format for a Signal Routine

Your action routine calls the signal routine using this format:

signal-procedure   fault-flag ,context ,signal-arguments 

fault-flag


OpenVMS usage: mask_longword
type: longword (unsigned)
access: read only
mechanism: by reference

Longword flag whose low-order bit determines whether the exception is to be signaled as a fault or as a trap. The fault-flag argument contains the address of this longword.

If the low-order bit of fault-flag is set to 1, the exception is signaled as a fault. If the low-order bit of fault-flag is set to 0, the exception is signaled as a trap; the current contents of the registers array are used. In either case, the current contents of PSL are used to set the exception PSL.

context


OpenVMS usage: context
type: unspecified
access: read only
mechanism: by reference

Context in which the new exception is to occur, as passed to your user action routine by LIB$DECODE_FAULT. The context argument is the address of this context value.

signal-arguments


OpenVMS usage: arg_list
type: longword (unsigned)
access: read only
mechanism: by reference, array reference

Signal arguments to be used. The signal-arguments argument is the address of an array of longwords that contains these signal arguments.

The first longword contains the number of following longwords; the remainder of the list contains signal names and arguments. Unlike the signal argument list passed to a condition handler, no PC or PSL is present.

Before the exception is signaled, the stack frames are unwound back to the original exception. You should be careful when causing a new signal that a loop of faults is not inadvertently generated. For example, the condition handler that called LIB$DECODE_FAULT will usually be called for the second signal. If the handler does not analyze the second signal as such, it may cycle through the identical path as for the first signal.

To resignal the current exception, have the user action routine return a value of SS$_RESIGNAL instead of calling the signal routine (unless you want previously called condition handlers to be called again).


Condition Values Returned

SS$_RESIGNAL Resignal condition to next handler. The exception described by signal-arguments was not an instruction fault handled by LIB$DECODE_FAULT. If LIB$DECODE_FAULT can process the fault, it does not return to its caller.

Condition Value Signaled

LIB$_INVARG Invalid argument to Run-Time Library. The instruction definition contained more than 16 operands or an operand definition contained an invalid data type or access code. This message is signaled after the stack frames have been unwound so that it appears to have been signaled from a routine that was called by the instruction that faulted.

Example

The following Fortran example implements a simple recovery scheme for floating underflow and overflow faults, replacing the result of the instruction with the correctly signed, smallest possible value for underflows or largest possible value for overflows.

 
C+ 
C  Example condition handler and user-action routine using 
C  LIB$DECODE_FAULT.  This example demonstrates the use of 
C  most of the features of LIB$DECODE_FAULT.  Its purpose 
C  is to handle floating underflow and overflow faults, 
C  replacing the result of the instruction with the correctly 
C  signed smallest possible value for underflows, or greatest 
C  possible value for overflows. 
C 
C  For simplicity, faults involving the POLYx instructions are 
C  not handled. 
C 
C*** 
C  FIXUP_RESULT is the condition handler enabled by the program 
C  desiring the fixup of overflows and underflows. 
C*** 
C- 
 
        INTEGER*4 FUNCTION FIXUP_RESULT(SIGARGS, MECHARGS) 
 
        IMPLICIT NONE 
        INCLUDE '($SSDEF)'              ! SS$_ symbols 
        INCLUDE '($LIBDCFDEF)'          ! LIB$DECODE_FAULT symbols 
        INTEGER*4 SIGARGS(1:*)          ! Signal arguments list 
        INTEGER*4 MECHARGS(1:*)         ! Mechanism arguments list 
 
C+ 
C This is a sample redefinition of MULH3 instruction. 
C- 
 
        BYTE OPTABLE(8) /'FD'X,'65'X,           ! MULH3 opcode 
        1                LIB$K_DCFOPR_RH,       ! Read H_floating 
        2                LIB$K_DCFOPR_RH,       ! Read H_floating 
        3                LIB$K_DCFOPR_WH,       ! Write H_floating 
        4                LIB$K_DCFOPR_END,      ! End of operands 
        5                'FF'X,'FF'X/           ! End of instructions 
 
        INTEGER*4 LIB$DECODE_FAULT      ! External function 
        EXTERNAL FIXUP_ACTION   ! Action routine to do the fixup 
 
 
C+ 
C       Determine if the exception is one we want to handle. 
C- 
 
 
        IF ((SIGARGS(2) .EQ. SS$_FLTOVF_F) .OR. 
        1   (SIGARGS(2) .EQ. SS$_FLTUND_F)) THEN 
 
C+ 
C         We think we can handle the fault.  Call 
C         LIB$DECODE_FAULT and pass it the signal arguments and 
C         the address of our action routine and opcode table. 
C- 
 
          FIXUP_RESULT = LIB$DECODE_FAULT (SIGARGS, 
        1   MECHARGS, %DESCR(FIXUP_ACTION),, OPTABLE) 
 
          RETURN 
        END IF 
 
C+ 
C       We can only get here if we couldn't handle the fault. 
C       Resignal the exception. 
C- 
 
        FIXUP_RESULT = SS$_RESIGNAL 
        RETURN 
        END 
 
C+ 
C  User action routine to handle the fault. 
C- 
 
        INTEGER*4 FUNCTION FIXUP_ACTION (OPCODE,INSTR_PC,PSL, 
        1                                REGISTERS,OP_COUNT, 
        2                                OP_TYPES,READ_OPS, 
        3                                WRITE_OPS,SIGARGS, 
        4                                SIGNAL_ROUT,CONTEXT, 
        5                                USER_ARG,ORIG_REGS) 
 
        IMPLICIT NONE 
        INCLUDE '($SSDEF)'              ! SS$_ definitions 
        INCLUDE '($PSLDEF)'             ! PSL$ definitions 
        INCLUDE '($LIBDCFDEF)'          ! LIB$DECODE_FAULT 
                                        ! definitions 
 
        INTEGER*4 OPCODE                ! Instruction opcode 
        INTEGER*4 INSTR_PC              ! PC of this instruction 
        INTEGER*4 PSL                   ! Processor status 
                                        ! longword 
        INTEGER*4 REGISTERS(0:15)       ! R0-R15 contents 
        INTEGER*4 OP_COUNT              ! Number of operands 
        INTEGER*4 OP_TYPES(1:*)         ! Types of operands 
        INTEGER*4 READ_OPS(1:*)         ! Addresses of read operands 
        INTEGER*4 WRITE_OPS(1:*)        ! Addresses of write operands 
        INTEGER*4 SIGARGS(1:*)          ! Signal argument list 
        INTEGER*4 SIGNAL_ROUT           ! Signal routine address 
        INTEGER*4 CONTEXT               ! Signal routine context 
        INTEGER*4 USER_ARG              ! User argument value 
        INTEGER*4 ORIG_REGS(0:15)       ! Original registers 
 
 
C+ 
C  Declare and initialize table of class codes for each of the 
C  "real" opcodes.  We'll index into this by the first byte of 
C  one-byte opcodes, the second byte of two-byte opcodes.  The 
C  class codes will be used in a computed GOTO (CASE).  The 
C  codes are: 
C               0 - Unsupported 
C               1 - ADD 
C               2 - SUB 
C               3 - MUL,DIV 
C               4 - ACB 
C               5 - CVT 
C               6 - EMOD 
C 
C  The class mainly determines how we compute the sign of the 
C  result, except for ACB. 
C- 
 
        BYTE INST_CLASS_TABLE(0:255) 
        DATA INST_CLASS_TABLE / 
        1       48*0,                                   ! 00-2F 
        2       0,0,0,5,0,0,0,0,0,0,0,0,0,0,0,0,        ! 30-3F 
        3       1,1,2,2,3,3,3,3,0,0,0,0,0,0,0,4,        ! 40-4F 
        4       0,0,0,0,6,0,0,0,0,0,0,0,0,0,0,0,        ! 50-5F 
        5       1,1,2,2,3,3,3,3,0,0,0,0,0,0,0,4,        ! 60-6F 
        6       0,0,0,0,6,0,5,0,0,0,0,0,0,0,0,0,        ! 70-7F 
        7       112*0,                                  ! 80-EF 
        8       0,0,0,0,0,0,5,5,0,0,0,0,0,0,0,0/        ! F0-FF 
 
C+ 
C  Table of operand sizes in 8-bit bytes, indexed by the 
C  datatype code contained in the OP_TYPES array.  Only floating 
C  types matter. 
C- 
 
        BYTE OP_SIZES(9) /0,0,0,0,0,4,8,8,16/ 
 
        INTEGER*4 LIB$EXTV              ! External function 
        INTEGER*4 RESULT_NEGATIVE       ! -1 if result negative, 
                                        ! 0 if positive 
        INTEGER*4 SIGN1,SIGN2,SIGN3     ! Signs of operands 
        INTEGER*4 INST_BYTE             ! Current opcode byte 
        INTEGER*4 INST_CLASS            ! Class of instruction 
                                        ! from table 
        INTEGER*4 OP_DTYPE              ! Datatype of operand 
        INTEGER*4 OP_SIZE               ! Size of operand in 
                                        ! 8-bit bytes 
        INTEGER*4 RESULT_OP             ! Position of result 
                                        ! in WRITE_OPS array 
        LOGICAL*4 OVERFLOW              ! TRUE if SS$_FLTOVF_F 
        LOGICAL*4 SMALLER               ! Function which 
                                        ! compares operands 
        PARAMETER ESCD = '0FD'X         ! First byte of G,H instructions 
 
        INTEGER*2 SMALL_F(2)            ! Smallest F_floating 
        DATA SMALL_F /'0080'X,0/ 
        INTEGER*2 SMALL_D(4)            ! Smallest D_floating 
        DATA SMALL_D /'0080'X,0,0,0/ 
        INTEGER*2 SMALL_G(4)            ! Smallest G_floating 
 
        DATA SMALL_G /'0010'X,0,0,0/ 
        INTEGER*2 SMALL_H(8)            ! Smallest H_floating 
        DATA SMALL_H /'0001'X,0,0,0,0,0,0,0/ 
        INTEGER*2 BIGGEST(8)            ! Biggest value (all datatypes) 
        DATA BIGGEST /'7FFF'X,7*'FFFF'X/ 
 
        INTEGER*4 SIGNAL_ARRAY(2)       ! Array for signalling new 
                                        ! exception 
C+ 
C 
C    NOTE:  Because the operands arrays contain the locations of 
C           the operands, rather than the operands themselves, 
C           we must call a routine using the %VAL function to 
C           "fool" the called routine into considering the 
C           contents of an operands array element as the address 
C           of an item.  This would not be necessary in a 
C           language that understood the concept of pointer 
C           variables, such as PASCAL. 
C 
C 
C  If FPD is set in the PSL, signal SS$_ROPRAND (reserved operand). In 
C  reality this shouldn't happen since none of the instructions we 
C  handle can set FPD, but do it as an example. 
C- 
 
        IF (BTEST(PSL,PSL$V_FPD)) THEN 
          SIGNAL_ARRAY(1) = 1           ! Count of signal arguments 
          SIGNAL_ARRAY(2) = SS$_ROPRAND ! Error status value 
          CALL SIGNAL_ROUT ( 
        1       1,                      ! Fault flag - signal as fault 
        2       SIGNAL_ARRAY,           ! Signal arguments array 
        3       CONTEXT)                ! Context as passed to us 
                                        ! Call will never return 
          END IF 
 
C+ 
C  Set OVERFLOW according to the exception type.  We assume that 
C  the only alternatives are SS$_FLTOVF_F and SS$_FLTUND_F. 
C- 
 
        OVERFLOW = (SIGARGS(2) .EQ. SS$_FLTOVF_F) 
 
C+ 
C  Determine the datatype of the instruction by that of its 
C  second operand, since that is always the type of the 
C  destination. 
C- 
 
        OP_DTYPE = IBITS(OP_TYPES(2),LIB$V_DCFTYP,LIB$S_DCFTYP) 
 
C+ 
C  Get the size of the datatype in words. 
C- 
 
        OP_SIZE = OP_SIZES (OP_DTYPE) 
 
C+ 
C  Determine the class of instruction and dispatch to the 
C  appropriate routine. 
C- 
 
 
        INST_BYTE = IBITS(OPCODE,0,8)   ! Get first byte 
        IF (INST_BYTE .EQ. ESCD) INST_BYTE = IBITS(OPCODE,8,8) 
        INST_CLASS = INST_CLASS_TABLE(INST_BYTE) 
        GO TO (1000,2000,3000,4000,5000,6000),INST_CLASS 
 
C+ 
C  If we get here, the instruction's entry in the 
C  INST_CLASS_TABLE is zero. This might happen if the instruction was 
C  a POLYx, or was some other unsupported instruction.  Resignal the 
C  original exception. 
C- 
 
        FIXUP_ACTION = SS$_RESIGNAL     ! Resignal condition to next handler 
        RETURN                          ! Return to LIB$DECODE_FAULT 
 
 
C+ 
C  1000 - ADDF2, ADDF3, ADDD2, ADDD3, ADDG2, ADDG3, ADDH2, ADDH3 
C 
C  Result's sign is the same as that of the first operand, 
C  unless this is an underflow, in which case the magnitudes of 
C  the values may change the sign. 
C- 
 
1000    RESULT_NEGATIVE = LIB$EXTV (15,1,%VAL(READ_OPS(1))) 
        IF (.NOT. OVERFLOW) THEN 
          IF (SMALLER(OP_SIZE,%VAL(READ_OPS(1)), 
        1                     %VAL(READ_OPS(2)))) 
        2   RESULT_NEGATIVE = .NOT. RESULT_NEGATIVE 
          END IF 
        GO TO 9000 
 
C+ 
C  2000 - SUBF2, SUBF3, SUBD2, SUBD3, SUBG2, SUBG3, SUBH2, SUBH3 
C 
C  Result's sign is the opposite of that of the first operand, 
C  unless this is an underflow, in which case the magnitudes of 
C  the values may change the sign. 
C- 
 
2000    RESULT_NEGATIVE = .NOT. LIB$EXTV (15,1,%VAL(READ_OPS(1))) 
        IF (.NOT. OVERFLOW) THEN 
          IF (SMALLER(OP_SIZE,%VAL(READ_OPS(1)), 
        1                     %VAL(READ_OPS(2)))) 
        2   RESULT_NEGATIVE = .NOT. RESULT_NEGATIVE 
          END IF 
        GO TO 9000 
 
C+ 
C  3000 - MULF2, MULF3, MULD2, MULD3, MULG2, MULG3, MULH2, MULH3, 
C         DIVF2, DIVF3, DIVD2, DIVD3, DIVG2, DIVG3, DIVH2, DIVH3, 
C 
C  If the signs of the first two operands are the same, then the 
C  result's sign is positive, if they are not it is negative. 
C- 
 
3000    SIGN1 = LIB$EXTV (15,1,%VAL(READ_OPS(1))) 
        SIGN2 = LIB$EXTV (15,1,%VAL(READ_OPS(2))) 
        RESULT_NEGATIVE = SIGN1 .XOR. SIGN2 
 
        GOTO 9000 
 
C+ 
C  4000 - ACBF, ACBD, ACBG, ACBH 
C 
C  The result's sign is the same as that of the second operand 
C  (addend), unless this is underflow, in which case the 
C  magnitudes of the addend and index may change the sign. 
C  We must also determine if the branch is to be taken. 
C- 
 
4000    SIGN2 = LIB$EXTV (15,1,%VAL(READ_OPS(2))) 
        RESULT_NEGATIVE = SIGN2 
        IF (.NOT. OVERFLOW) THEN 
          IF (SMALLER(OP_SIZE,%VAL(READ_OPS(2)), 
        1                     %VAL(READ_OPS(3)))) 
        2   RESULT_NEGATIVE = .NOT. RESULT_NEGATIVE 
          END IF 
 
C+ 
C  If this is overflow, then the branch is not taken, since the 
C  result is always going to be greater or equal in magnitude 
C  to the limit, and will be the correct sign.  If underflow, 
C  the branch is ALMOST always taken.  The only case where the 
C  branch might not be taken is when the result is exactly 
C  equal to the limit.  For this example, we are going to ignore 
C  this exceptional case. 
C- 
 
        IF (.NOT. OVERFLOW) 
        1  REGISTERS(15) = READ_OPS(4)  ! Branch destination 
        GO TO 9000 
 
C+ 
C  5000 - CVTDF, CVTGF, CVTHF, CVTHD, CVTHG 
C 
C  Result's sign is the same as that of the first operand. 
C- 
 
5000    RESULT_NEGATIVE = LIB$EXTV (15,1,%VAL(READ_OPS(1))) 
        GO TO 9000 
 
C+ 
C  6000 - EMODF, EMODD, EMODG, EMODH 
C 
C  If the signs of the first and third operands are the same, then the 
C  result's sign is positive, else it is negative. 
C- 
 
6000    SIGN1 = LIB$EXTV (15,1,%VAL(READ_OPS(1))) 
        SIGN2 = LIB$EXTV (15,1,%VAL(READ_OPS(3))) 
        RESULT_NEGATIVE = SIGN1 .XOR. SIGN2 
        GOTO 9000 
 
C+ 
C  All code paths merge here to store the result value.  We also 
C  set the PSL appropriately.  First, determine which operand is 
C  the result. 
C- 
 
 
9000    RESULT_OP = OP_COUNT 
        IF (INST_CLASS .EQ. 4) 
        1  RESULT_OP = RESULT_OP - 1    ! ACBx 
 
C+ 
C       Select result based on datatype and exception type. 
C- 
 
        IF (OVERFLOW) THEN 
          CALL LIB$MOVC3 (OP_SIZE,BIGGEST,%VAL(WRITE_OPS(RESULT_OP))) 
        ELSE 
          GO TO (9100,9200,9300,9400), OP_DTYPE-(LIB$K_DCFTYP_F-1) 
 
C+ 
C         Should never get here.  Resignal original exception. 
C- 
 
          FIXUP_ACTION = SS$_RESIGNAL 
          RETURN 
 
C+ 
C  9100 - F_floating result 
C- 
 
9100      CALL LIB$MOVC3 (OP_SIZE,SMALL_F,%VAL(WRITE_OPS(RESULT_OP))) 
          GOTO 9500 
 
C+ 
C  9200 - D_floating result 
C- 
 
9200      CALL LIB$MOVC3 (OP_SIZE,SMALL_D,%VAL(WRITE_OPS(RESULT_OP))) 
          GOTO 9500 
 
C+ 
C  9300 - G_floating result 
C- 
 
9300      CALL LIB$MOVC3 (OP_SIZE,SMALL_G,%VAL(WRITE_OPS(RESULT_OP))) 
          GOTO 9500 
 
C+ 
C  9400 - H_floating result 
C- 
 
9400      CALL LIB$MOVC3 (OP_SIZE,SMALL_H,%VAL(WRITE_OPS(RESULT_OP))) 
          GOTO 9500 
 
9500    END IF 
 
C+ 
C  Modify the PSL to reflect the stored result.  If the result was 
C  negative, set the N bit.  Clear the V (overflow) and Z (zero) bits. 
C  If the instruction was an ACBx, leave the C (carry) bit unchanged, 
C  otherwise clear it. 
C- 
 
        IF (RESULT_NEGATIVE) THEN 
          PSL = IBSET (PSL,PSL$V_N)     ! Set N bit 
        ELSE 
 
          PSL = IBCLR (PSL,PSL$V_N)     ! Clear N bit 
        END IF 
        PSL = IBCLR (PSL,PSL$V_V)       ! Clear V bit 
        PSL = IBCLR (PSL,PSL$V_Z)       ! Clear Z bit 
        IF (INST_CLASS .NE. 4) 
        1  PSL = IBCLR (PSL,PSL$V_C)    ! Clear C bit if not ACBx 
 
C+ 
C  Set the sign of result. 
C- 
 
        IF (RESULT_NEGATIVE) 
        1  CALL LIB$INSV (1,15,1,%VAL(WRITE_OPS(RESULT_OP))) 
C+ 
C  Fixup is complete.  Return to LIB$DECODE_FAULT. 
C- 
 
        FIXUP_ACTION = SS$_CONTINUE 
        RETURN 
        END 
 
C+ 
C Function which compares two floating values.  It returns .TRUE. if 
C the first argument is smaller in magnitude than the second. 
C- 
 
        LOGICAL*4 FUNCTION SMALLER(NBYTES,VAL1,VAL2) 
 
        INTEGER*4 NBYTES                ! Number of bytes in values 
        INTEGER*2 VAL1(*),VAL2(*)       ! Floating values to compare 
        INTEGER*4 WORDA,WORDB 
 
        SMALLER = .TRUE.                ! Initially return true 
 
C+ 
C       Zero extend to a longword for unsigned compares. 
C       Compare first word without sign bit. 
C- 
 
        WORDA = IBCLR(ZEXT(VAL1(1)),15) 
        WORDB = IBCLR(ZEXT(VAL2(1)),15) 
        IF (WORDA .LT. WORDB) RETURN 
 
        DO I=2,NBYTES/2 
        WORDA = ZEXT(VAL1(I)) 
        WORDB = ZEXT(VAL2(I)) 
        IF (WORDA .LT. WORDB) RETURN 
        END DO 
 
        SMALLER = .FALSE.       ! VAL1 not smaller than VAL2 
        RETURN 
        END 
 
      


Previous Next Contents Index

[Site home] [Send comments] [Help with this site] [How to order documentation] [OpenVMS site] [Compaq site]
[OpenVMS documentation]

Copyright © Compaq Computer Corporation 1998. All rights reserved.

Legal
5932PRO_010.HTML