Instruction methods for performing data formatting while moving data between memory and a vector register file

In a computer controlled graphics system, a method of loading a data segment into a register file, said register file comprising a plurality of registers with individual registers being 128 bits wide, said method comprising the steps of:

• a) processing a single load instruction containing an initial memory location, said single load instruction performing the steps of:
• a1) identifying said initial memory location within a memory unit of said computer controlled graphics system;
• a2) simultaneously accessing eight consecutive bytes of said data segment starting at said memory location;
• and
• a3) simultaneously loading said eight consecutive bytes into consecutive locations of a same register of said register file, wherein said consecutive locations of said register are each 16-bits wide.

In a computer controlled graphics system, a method of loading a data segment into a register file, said register file comprising a plurality of registers with individual registers being 128 bits wide, said method comprising the steps of:

• a) processing a single load instruction containing an initial memory location, said single load instruction performing the steps of:
• a1) identifying said initial memory location within a memory unit of said computer controlled graphics system;
• a2) simultaneously accessing eight alternate bytes of said data segment starting at said memory location with a byte offset value;
• and
• a3) simultaneously loading said eight alternate bytes into consecutive locations of a same register of said register file;
• wherein said consecutive locations of said same register are each 16-bits wide and wherein steps a2) and a3) occur within a single clock cycle.

In a computer controlled graphics system, a method of loading a data segment into a register file, said register file comprising a plurality of registers with individual registers being 128 bits wide, said method comprising the steps of:

• a) processing a single load instruction containing an initial memory location, said single load instruction performing the steps of:
• a1) identifying said initial memory location within a memory unit of said computer controlled graphics system;
• a2) simultaneously accessing every fourth byte of said data segment starting at said memory location with a byte offset value;
• and
• a3) simultaneously loading said every fourth byte of step a2) into four consecutive locations of a same register of said register file;
• wherein said four consecutive locations of said register are each 16-bits wide and wherein steps a2) and a3) occur within a single clock cycle.

In a computer controlled graphics system, a method of loading a data segment into a register file, said register file comprising a plurality of registers with individual registers being 128 bits wide, said method comprising the steps of:

• a) processing a single load instruction containing an initial memory location, said single load instruction performing the steps of:
• a1) identifying said initial memory location within a memory unit of said computer controlled graphics system;
• a2) simultaneously accessing four 16-bit shortwords of said data segment starting at said memory location with a byte offset value;
• and
• a3) simultaneously loading said four 16-bit shortwords into four consecutive locations of a same register of said register file;
• wherein said consecutive locations of said same register are each 16-bits wide and wherein steps a2) and a3) occur within a single clock cycle.

In a computer controlled graphics system including a processor, a memory unit, and a register file of said processor, a method of transposing data of an original matrix stored in said memory unit, said original matrix comprising N rows and N columns of y-bit data portions, said method comprising the steps of:

• a) executing a single load instruction to simultaneously access N data portions of a row of said original matrix from said memory unit and to simultaneously load said N data portions of said row into respective y-bit locations of N consecutive registers of said register file commencing at an initial register;
• b) executing said step a) N times with N different initial registers, one execution for each row of said N rows of said original matrix, said step b) producing an intermediate matrix stored in said register file;
• c) executing a single store instruction to perform a left shift of N y-bit locations of a register of said register file to arrive at N left shifted y-bit values and to simultaneously load said N left shifted y-bit values into a row of memory unit;
• and
• d) executing said step c) for each of said N registers used by step b) wherein a different left shift amount is performed and a different row of said memory unit is used for each execution, said step d) producing a transposed matrix of N rows stored within said memory unit, said transposed matrix being a transposition of said original matrix.

In a computer controlled graphics system including a processor, a memory unit, and a register file of said processor, a method of transposing data of an original matrix stored in said register file, said original matrix comprising N rows and N columns of y-bit data portions, said method comprising the steps of:

• a) executing a single store instruction to simultaneously access N data portions of N different y-bit locations of N different registers of said register file, commencing at an initial register, and to simultaneously store said N data portions into N consecutive y-bit locations of a row of memory of said memory unit;
• b) executing said step a) N times with N different initial registers, said step b) producing an intermediate N by N matrix in said memory unit with each execution of step a) filling a different row of said intermediate matrix;
• c) executing a single load instruction to simultaneously access N data portions of a row from said intermediate matrix of said memory unit and to simultaneously load said N data portions of said row into respective y-bit locations of N consecutive registers of said register file commencing at an initial register;
• and
• d) executing said step c) for each row of said intermediate matrix, said step d) producing a transposed matrix stored within said register file, said transposed matrix being a transposition of said original matrix.