Floating-point processor provided with high-speed detector of overflow and underflow exceptional conditions

Claim 2

A processor comprising: an exception detector receiving input floating-point data on which a rounding operation is to be performed and detecting, without using output floating-point data which is obtained by performing a rounding operation on said input floating-point data, whether or not an overflow or an underflow exception occurs in said output floating-point data, said exception detector including first means for producing first data indicating whether a rounding operation to be executed is a rounding operation by raising or a rounding operation by truncating, second means for unconditionally performing a rounding operation by raising a mantissa part of said input floating-point data to provide a modified mantissa part having overflow data, third means for evaluating an exponent part of said input floating-point data to produce second data taking a first state when said exponent part is a valid maximum number and taking a second state when said exponent part is zero, and fourth means, responsive to said first data, said second data and said overflow data, for producing detection data indicating whether or not said overflow or said underflow exception occurs in said output floating-point data subjected to said rounding operation.

Claim 3

A processor as claimed in claim 1, wherein at least said first, second, and third means operate substantially in parallel to one another.

Claim 4

A processor as claimed in claim 1, wherein said fourth means produces detection data representing occurrence of said overflow exception in said output floating-point data when said first data indicates that said rounding operation to be executed is said rounding operation by raising and when said second data takes said first state and when said overflow data indicates that an overflow is produced in said modified mantissa part.

Claim 5

A processor as claimed in claim 1, wherein said fourth means produces detection data indicating occurrence of said underflow exception in said output floating-point data when said second data takes said second state and when said first data indicates that said rounding operation to be executed is said rounding operation by truncation or said overflow data indicates that no overflow is produced in said modified mantissa part.

Claim 6

A processor as claimed in claim 1, wherein said first means includes means for temporarily storing mode data for designating a rounding operation and means for determining a content of said first data in response to said mode data and at least one part of said mantissa part.

Claim 7

A floating-point data processor comprising: an exception detector receiving input floating-point data on which a rounding operation is to be performed and detecting whether or not an overflow or an underflow exception occurs in output floating-point data which is obtained by performing a rounding operation on said input floating-point data, said exception detector including storage means for temporarily storing mode data for designating a rounding operation; first means responsive to said mode data and to at least one part of a mantissa part of said input floating-point data, said first data taking a first state when a rounding operation by raising is determined to be performed and a second state when a rounding operation by truncating is determined to be performed; second means for incrementing said mantissa part of said input floating-point data to produce a modified mantissa part having overflow data, said overflow data taking a third state when an overflow occurs in said modified mantissa part and a fourth state when no overflow occurs in said modified mantissa part; third means for evaluating an exponent part of said input floating-point data to produce second data, said second data taking a fifth state when said exponent part is a valid maximum number, a sixth state when said exponent part is zero, a seventh state when said exponent part is larger than said valid maximum number and an eighth state when said exponent part is a negative number; a first terminal from which a first signal is generated to indicate occurrence of said overflow exception in said output floating-point data; a second terminal from which a second signal is generated to indicate occurrence of said underflow exception in said output floating-point data; and a logic gate circuit supplied with said first, second and third data for detecting whether or not said overflow or said underflow exception occurs in said output floating-point data, said logic gate circuit generating said first signal at said first terminal when said overflow exception occurs and said second signal at said second terminal when said underflow exception occurs.

Claim 8

A floating-point data processor comprising: an exception detector receiving input floating-point data on which a rounding operation is to be performed and detecting whether or not an overflow or an underflow exception occurs in output floating-point data which is obtained by performing a rounding operation on said input floating-point data, said exception detector including storage means for temporarily storing mode data for designating a rounding operation; first means responsive to said mode data and to at least one part of a mantissa part of said input floating-point data, said first data taking a first state when a rounding operation by raising is determined to be performed and a second state when a rounding operation by truncating is determined to be performed; second means for incrementing said mantissa part of said input floating-point data to produce a modified mantissa part having overflow data, said overflow data taking a third state when an overflow occurs in said modified mantissa part and a fourth state when no overflow occurs in said modified mantissa part; third means for evaluating an exponent part of said input floating-point data to produce second data, said second data taking a fifth state when said exponent part is a valid maximum number, a sixth state when said exponent part is zero, a seventh state when said exponent part is larger than said valid maximum number and an eighth state when said exponent part is a negative number; a first terminal from which a first signal is generated to indicate occurrence of said overflow exception in said output floating-point data; a second terminal from which a second signal is generated to indicate occurrence of said underflow exception in said output floating-point data; and a logic gate circuit supplied with said first, second and third data for detecting whether or not said overflow or said underflow exception occurs in said output floating-point data, said logic gate circuit generating said first signal at said first terminal when said overflow exception occurs and said second signal at said second terminal when said underflow exception occurs.

Claim 9

A processor as claimed in claim 6, wherein said first signal is generated by said logic gate circuit when said first data takes said first state and said overflow data takes said third state and said second data takes said fifth state, or when said second data takes said seventh state.

Claim 10

A processor as claimed in claim 6, wherein said second signal is generated by said logic gate circuit when said second data takes said sixth state and said first data takes said second state or said overflow data takes said fourth state, or when said second data takes said eighth state.

Claim 11

A floating-point data processor comprising: an exception detector receiving input floating-point data on which a rounding operation is to be performed and detecting whether or not an overflow or an underflow exception occurs in output floating-point data which is obtained by performing a rounding operation on said input floating-point data, said exception detector including storage means for temporarily storing mode data for designating a rounding operation; first means responsive to said mode data and at least one part of a mantissa part of said input floating-point data for producing first data for determining a rounding operation to be performed on said input floating-point data, said first data taking a first state when a rounding operation by raising is determined to be performed and a second state when a rounding operation by truncating is determined to be performed; second means for incrementing said mantissa part of said input floating-point data to produce a modified mantissa part having overflow data, said overflow data taking a third state when an overflow occurs in said modified mantissa part and a fourth state when no overflow occurs in said modified mantissa part; third means for evaluating an exponent part of said input floating-point data to produce second data, said second data taking a fifth state when said exponent part is a valid maximum number, a sixth state when said exponent part is zero, a seventh state when said exponent part is larger than said valid maximum number and an eighth state when said exponent part is a negative number; a first terminal from which a first signal is generated to indicate occurrence of said overflow exception in said output floating-point data; a second terminal from which a second signal is generated to indicate occurrence of said underflow exception in said output floating-point data; a memory circuit having a plurality of storage locations; means for combining said first, second and third data to supply combined data to said memory circuit as address data, said memory circuit reading out contents from a storage location accessed by said address data; means responsive to a content read out of said memory circuit for generating said first signal at said first terminal when said content represents the occurrence of said overflow exception; and means responsive to a content read out of said memory circuit for generating said second signal at said second terminal when said content represents occurrence of said underflow exception.

Claim 12

A floating-point data processor comprising: an exception detector receiving input floating-point data on which a rounding operation is to be performed and detecting whether or not an overflow or an underflow exception occurs in output floating-point data which is obtained by performing a rounding operation on said input floating-point data, said exception detector including storage means for temporarily storing mode data for designating a rounding operation; first means responsive to said mode data and at least one part of a mantissa part of said input floating-point data for producing first data for determining a rounding operation to be performed on said input floating-point data, said first data taking a first state when a rounding operation by raising is determined to be performed and a second state when a rounding operation by truncating is determined to be performed; second means for incrementing said mantissa part of said input floating-point data to produce a modified mantissa part having overflow data, said overflow data taking a third state when an overflow occurs in said modified mantissa part and a fourth state when no overflow occurs in said modified mantissa part; third means for evaluating an exponent part of said input floating-point data to produce second data, said second data taking a fifth state when said exponent part is a valid maximum number, a sixth state when said exponent part is zero, a seventh state when said exponent part is larger than said valid maximum number and an eighth state when said exponent part is a negative number; a first terminal from which a first signal is generated to indicate occurrence of said overflow exception in said output floating-point data; a second terminal from which a second signal is generated to indicate occurrence of said underflow exception in said output floating-point data; a memory circuit having a plurality of storage locations; means for combining said first, second and third data to supply combined data to said memory circuit as address data, said memory circuit reading out contents from a storage location accessed by said address data; means responsive to a content read out of said memory circuit for generating said first signal at said first terminal when said content represents the occurrence of said overflow exception; and means responsive to a content read out of said memory circuit for generating said second signal at said second terminal when said content represents occurrence of said underflow exception.