Automatic identification and generation of non-temporal store and load operations in a dynamic optimization environment

A system comprising:

• a processor;
• a memory coupled to the processor;
• a profiler to:
• identify a frequently-executed region of code during runtime execution of the code by the processor;
• generate initial profiling code for the frequently-executed region of code;
• cause the initial profiling code to execute for a minimum number of processing cycles;
• and
• identify at least one replacement candidate store instruction that stores a value, wherein the value is not read by the frequently-executed region of code during execution of the initial profiling code, wherein the profiler further is to:
• for each store instruction in the frequently-executed region of code, insert, into the initial profiling code, the store instruction and a first instruction to record information about the store instruction;
• and
• for each load instruction in the frequently-executed region of code, insert, into the initial profiling code, the load instruction and a second instruction to record information about the load instruction;
• and
• a runtime code generator to:
• generate optimized code for the frequently-executed region of code, wherein to generate the optimized code comprises to replace the at least one replacement candidate store instruction with a non-temporal store instruction;
• and
• cause the optimized code to execute instead of the frequently-executed region of code during subsequent runtime execution, wherein the runtime code generator is further to:
• cause the first instruction to execute to record the information about the store instruction, wherein execution of the first instruction causes the system to:
• generate a unique storage identifier for the store instruction;
• and
• record an association between the unique storage identifier for the store instruction and a storage address in which the store instruction stores data;
• and
• cause the second instruction to execute to record the information about the load instruction, wherein execution of the second instruction causes the system to:
• determine whether a previous store instruction accesses a load address from which the load instruction loads data;
• and
• if the previous store instruction accesses the load address, mark the unique storage identifier for the previous store instruction as a temporal store instruction.

A method comprising:

• identifying a frequently-executed region of code during runtime execution of the code;
• generating initial profiling code for the frequently-executed region of code, wherein generating the initial profiling code for the frequently-executed region of code comprises:
• for each store instruction in the frequently-executed region of code, inserting, into the initial profiling code, the store instruction and a first instruction to record information about the store instruction;
• for each load instruction in the frequently-executed region of code, inserting, into the initial profiling code, the load instruction and a second instruction to record information about the load instruction;
• causing the initial profiling code to be executed for a given number of iterations, wherein causing the initial profiling code to execute comprises:
• causing the first instruction to execute to record the information about the store instruction, wherein execution of the first instruction comprises:
• generating a unique storage identifier for the store instruction;
• and
• recording an association between the unique storage identifier for the store instruction and a storage address in which the store instruction stores data;
• and
• causing the second instruction to execute to record the information about the load instruction, wherein execution of the second instruction comprises:
• determining whether a previous store instruction accesses a load address from which the load instruction loads data;
• and
• if the previous store instruction accesses the load address, marking the unique storage identifier for the previous store instruction as a temporal store instruction;
• identifying at least one replacement candidate store instruction that stores a stored value, wherein the stored value is not read by the frequently-executed region of code during execution of the initial profiling code;
• generating optimized code for the frequently-executed region of code, wherein generating the optimized code comprises replacing the at least one replacement candidate store instruction with a non-temporal store instruction;
• and
• causing the optimized code to be executed instead of the frequently-executed region of code during subsequent runtime execution.