HP OpenVMS Systems Documentation

The VAX vector architecture provides instructions for operating on F_floating, D_floating, and G_floating operand formats. The floating-point arithmetic instructions are add, subtract, compare, multiply, and divide. Data conversion instructions are provided to convert operands between D_floating, G_floating, F_floating, and longword integer.

Rounding is performed using standard VAX rounding rules. The accuracy of the vector floating-point instructions matches that of the scalar floating-point instructions. Refer to the section on floating-point instructions in the VAX Architecture Reference Manual for more information.

10.13.1 Vector Floating-Point Exception Conditions

All vector floating-point exception conditions occur asynchronously with respect to the scalar processor. These exception conditions do not interrupt the scalar processor. If the exception condition is enabled, then the exception condition type and destination register number are recorded in the Vector Arithmetic Exception Register (VAER), and a reserved operand in the format of the instruction's data type is written into the destination register element. Encoded in this reserved operand is the exception condition type. After recording the exception and writing the appropriate result into the destination register element, the instruction encountering the exception continues executing to completion.

If a vector convert floating to integer instruction encounters a source element that is a reserved operand, an UNPREDICTABLE result rather than a reserved operand is written into the destination register element.

Figure 10-13 shows the encoding of the reserved operand that is written for vector floating-point exceptions. Consistent with the definition of a reserved operand, the sign bit (bit <15>) is one and the exponent (bits <14:7> for F_floating and D_floating, and bits <14:4> for G_floating) is zero. When the reserved operand is written in F_floating or D_floating format, bits <6:4> are also zero. The exception condition type (ETYPE) is encoded in bits <3:0>, as shown in Table 10-16. If a reserved operand is divided by zero, both ETYPE bits may be set. The state of all other bits in the result (denoted by shading) is UNPREDICTABLE.

If the floating underflow exception condition is suppressed by cntrl<EXC>, a zero result is written to the destination register element and no further action is taken. Floating overflow, floating divide by zero, and floating reserved operand are always enabled.

Figure 10-13 Encoding of the Reserved Operand

This section describes VAX vector architecture floating-point instructions.

Vector Floating Add

Format

vector + vector:

{VVADDF|VVADDD|VVADDG} [/U[0|1]] Va, Vb, Vc

scalar + vector:

{VSADDF|VSADDD|VSADDG [/U[0|1]] scalar, Vb, Vc

Format

vector + vector:

opcode cntrl.rw

scalar + vector (F_floating):

opcode cntrl.rw, addend.rl

scalar + vector (D_ and G_floating):

opcode cntrl.rw, addend.rq

Opcodes

vector_control_word

Exceptions

floating overflow
floating reserved operand
floating underflow

Description

The source addend or vector operand Va is added, elementwise, to vector register Vb and the sum is written to vector register Vc. The length of the vector is specified by the Vector Length Register (VLR).

In VxADDF, only bits <31:0> of each vector element participate in the operation. Bits <63:32> of the destination vector elements are UNPREDICTABLE.

If a floating underflow occurs when cntrl<EXC> is set or if a floating overflow or floating reserved operand occurs, an encoded reserved operand is stored as the result and the exception condition type and destination register number are recorded in the Vector Arithmetic Exception Register (VAER). The vector operation is then allowed to complete. If cntrl<EXC> is clear, zero is written to the destination element when an exponent underflow occurs and no other action is taken.

Vector Floating Compare

Format

vector--vector:

{VVGTRF|VVGTRD|VVGTRG|VVEQLF|VVEQLD| VVEQLG|VVLSSF|VVLSSD|VVLSSG|VVLEQF|VVLEQD| VVLEQG|VVNEQF|VVNEQD|VVNEQG|VVGEQF|VVGEQD|VVGEQG} [/U[0|1]] Va, Vb

scalar--vector:

{VSGTRF|VSGTRD|VSGTRG|VSEQLF|VSEQLD| VSEQLG|VSLSSF|VSLSSD|VSLSSG|VSLEQF|VSLEQD| VSLEQG|VSNEQF|VSNEQD|VSNEQG|VSGEQF|VSGEQD|VSGEQG} [/U[0|1]] src, Vb

Architecture

Format

vector--vector:

opcode cntrl.rw

scalar--vector (F_floating):

opcode cntrl.rw, src.rl

scalar--vector (D_ and G_floating):

opcode cntrl.rw, src.rq

Opcodes

vector_control_word

The condition being tested is determined by cntrl<2:0>, as follows:

Exceptions

floating reserved operand

Description

The scalar or vector operand Va is compared, elementwise, with vector register Vb. The length of the vector is specified by the Vector Length Register (VLR). For each element comparison, if the specified relationship is true, the Vector Mask Register bit (VMR<i>) corresponding to the vector element is set to one, otherwise it is cleared. If cntrl<MOE> is set, VMR bits corresponding to elements that do not match cntrl<MTF> are left unchanged. VMR bits beyond the vector length are left unchanged. If an element being compared is a reserved operand, VMR<i> is UNPREDICTABLE. In VxCMPF, only bits <31:0> of each vector element participate in the operation.

If a floating reserved operand exception occurs, the exception condition type is recorded in the Vector Arithmetic Exception Register (VAER) and the vector operation is allowed to complete.

Note that for this instruction, no bits are set in the VAER destination register mask when an exception occurs.

Vector Convert

Format

{VVCVTLF|VVCVTLD|VVCVTLG|VVCVTFL|VVCVTRFL| VVCVTFD|VVCVTFG|VVCVTDL|VVCVTDF| VVCVTRDL|VVCVTGL|VVCVTGF|VVCVTRGL} [/U[0|1]] Vb, Vc

Format

opcode cntrl.rw

Opcodes

vector_control_word

Cntrl<11:8> specifies the conversion to be performed, as follows:

Exceptions

floating overflow
floating reserved operand
floating underflow
integer overflow

Description

The vector elements in vector register Vb are converted and results are written to vector register Vc. Cntrl<11:8> specifies the conversion to be performed. The length of the vector is specified by the Vector Length Register (VLR). Bits <63:32> of Vc are UNPREDICTABLE for instructions that convert from D_floating or G_floating to F_floating or longword. When CVTRGL, CVTRDL, and CVTRFL round, the rounding is done in sign magnitude, before conversion to two's complement.

If an integer overflow occurs when cntrl<EXC> is set, the low-order 32 bits of the true result are written to the destination element as the result, and the exception condition type and destination register number are recorded in the Vector Arithmetic Exception Register (VAER). The vector operation is then allowed to complete. If integer overflow occurs when cntrl<EXC> is clear, the low-order 32 bits of the true result are written to the destination element, and no other action is taken.

For vector convert floating to integer, where the source element is a reserved operand, the value written to the destination element is UNPREDICTABLE. In addition, the exception type and destination register number are recorded in the VAER. The vector operation is then allowed to complete.

For vector convert floating to floating instructions, if floating underflow occurs when cntrl<EXC> is clear, zero is written to the destination element, and no other action is taken. The vector operation is then allowed to complete.

For vector convert floating to floating instructions, if floating underflow occurs with cntrl<EXC> set or if a floating overflow or reserved operand occurs, an encoded reserved operand is written to the destination element, and the exception condition type and destination register number are recorded in the VAER. The vector operation is then allowed to complete.

Vector Floating Divide

Format

vector/vector:

{VVDIVF|VVDIVD|VVDIVG} [/U[0|1]] Va, Vb, Vc

scalar/vector:

{VSDIVF|VSDIVD|VSDIVG} [/U[0|1]] scalar, Vb, Vc

Format

vector/vector:

opcode cntrl.rw

scalar/vector (F_floating):

opcode cntrl.rw, divd.rl

scalar/vector (D_ and G_floating):

opcode cntrl.rw, divd.rq

Opcodes

vector_control_word

Exceptions

floating divide by zero
floating overflow
floating reserved operand
floating underflow

Description

The scalar dividend or vector register Va is divided, elementwise, by the divisor in vector register Vb and the quotient is written to vector register Vc. The length of the vector is specified by the Vector Length Register (VLR).

In VxDIVF, only bits <31:0> of each vector element participate in the operation; bits <63:32> of the destination vector elements are UNPREDICTABLE.

If a floating underflow occurs when cntrl<EXC> is set or if a floating overflow, divide by zero, or reserved operand occurs, an encoded reserved operand is stored as the result and the exception condition type and destination register number are recorded in the Vector Arithmetic Exception Register (VAER). The vector operation is then allowed to complete. If cntrl<EXC> is clear, zero is written to the destination element when an exponent underflow occurs and no other action is taken.

Vector Floating Multiply

Format

vector * vector:

{VVMULF|VVMULD|VVMULG} [/U[0|1]] Va, Vb, Vc

scalar * vector:

{VSMULF|VSMULD|VSMULG} [/U[0|1]] scalar, Vb, Vc

Format

vector * vector:

opcode cntrl.rw

scalar * vector (F_floating):

opcode cntrl.rw, mulr.rl

scalar * vector (D_ and G_floating):

opcode cntrl.rw, mulr.rq

Opcodes

vector_control_word

Exceptions

floating overflow
floating reserved operand
floating underflow

Description

The multiplicand in vector register Vb is multiplied, elementwise, by the scalar multiplier or vector operand Va and the product is written to vector register Vc. The length of the vector is specified by the Vector Length Register (VLR).

In VxMULF, only bits <31:0> of each vector element participate in the operation. Bits <63:32> of the destination vector elements are UNPREDICTABLE.

If a floating underflow occurs when cntrl<EXC> is set or if a floating overflow or reserved operand occurs, an encoded reserved operand is stored as the result and the exception condition type and destination register number are recorded in the Vector Arithmetic Exception Register (VAER). The vector operation is then allowed to complete. If cntrl<EXC> is clear, zero is written to the destination element when an exponent underflow occurs and no other action is taken.

Vector Floating Subtract

Format

vector--vector:

{VVSUBF|VVSUBD|VVSUBG} [/U[0|1]] Va, Vb, Vc

scalar--vector:

{VSSUBF|VSSUBD|VSSUBG} [/U[0|1]] scalar, Vb, Vc

Format

vector--vector:

opcode cntrl.rw

scalar--vector (F_floating):

opcode cntrl.rw, min.rl

scalar--vector (D_ and G_floating):

opcode cntrl.rw, min.rq

Opcodes

vector_control_word

Exceptions

floating overflow
floating reserved operand
floating underflow

Description

Vector register Vb is subtracted, elementwise, from the scalar minuend or vector register Va and the difference is written to vector register Vc. The length of the vector is specified by the Vector Length Register (VLR).

In VxSUBF, only bits <31:0> of each vector element participate in the operation; bits <63:32> of the destination vector elements are UNPREDICTABLE.

If a floating underflow occurs when cntrl<EXC> is set or if a floating overflow or reserved operand occurs, an encoded reserved operand is stored as the result and the exception condition type and destination register number are recorded in the Vector Arithmetic Exception Register (VAER). The vector operation is then allowed to complete. If cntrl<EXC> is clear, zero is written to the destination element when an exponent underflow occurs and no other action is taken.