A uniprocessor sys accesses memory over a 3 write-back (as opposed to th data cache employing a write-allocate a data cache perates contains 65536 lines each through mode, uses true takes clock of which is bytes in size. Each data cache access perform a cycles to detect cache miss, or to perform a cache read or back to cache ding a memory block into cache a cache line item that takes 400 CPU clock cycles. Recall that for a cache the data item is needed is read in parallel with loading the memory block containing the after a into cache. The operating system flushes the of cache program terminates. The code shown below reads and updates each of the 134217728 four-byte elements in an integer array. The address of the array in memory is 0x100840A0. lui $8,0x1008 point to first element to be processed ori $8,0x40A0 loop:
lui $4,0x0800 number of elements 134217728 0x08000000)
lw $12,0 ($8) get next element
addi $4,$4,-1 decrement loop control variable
sub $12, $0,$12 negate by subtracting from o
addi $8,$8,4 point to next element
sw $12,-4($8) update element that was read
bgez $4, loop repeat if more elements remain nop
In an attempt to speedup the processing of the array, the code is executed on an 8-core system in which each core has a separate data cache identical to the data cache for the uniprocessor. A speedup factor of 8 for the 8-core system compared to the uniprocessor would correspond to what is called "linear" speedup.
What would be the actual speedup provided by the 8-core system based on just the time required to read and update the array elements, ignoring the time required by the instructions that do not reference memory? Each of the 8 cores executes a separate copy of the code but with the appropriate starting address and loop limit that would correspond to 1/8th of the array. Assume that all data caches are initially empty.