A processor with a frequency of 2GHz executes the following code:
for (i=0;i<4096;i++) {
s = s + X * Y
}
Assume that s is in a register (8 bytes), vectors X and Y are aligned to block level in memory. All elements of X
and Y are 8 bytes long. X and Y hold FP values.
The processor has a data cache (L1) of 32 KBytes, two‐way set associative, and block size of 64 Bytes. The
miss penalty is 128 cycles (time to access the main memory). On a miss, the CPU stalls the execution of
instructions until the miss is served (the data is sent to the CPU). The instruction cache does not have any
miss. Assume vectors X and Y are NOT loaded in the L1 when the execution starts.
The compiler translates the code to 8 instructions, two of which are memory accesses. The average CPI when
we hit in the cache is 1.
Question 1: Compute the total number of misses
Question 2: Compute the misses per instruction
Question 3: Taking into account the misses, compute the CPI, MIPS and MFLOPS.
Now assume we add a second level of cache (L2) and assume that vectors X and Y are in the L2 when the
execution starts.
Question 4: Compute the miss penalty for the L1 to double the performance.
Hope someone can help me thanks
for (i=0;i<4096;i++) {
s = s + X * Y
}
Assume that s is in a register (8 bytes), vectors X and Y are aligned to block level in memory. All elements of X
and Y are 8 bytes long. X and Y hold FP values.
The processor has a data cache (L1) of 32 KBytes, two‐way set associative, and block size of 64 Bytes. The
miss penalty is 128 cycles (time to access the main memory). On a miss, the CPU stalls the execution of
instructions until the miss is served (the data is sent to the CPU). The instruction cache does not have any
miss. Assume vectors X and Y are NOT loaded in the L1 when the execution starts.
The compiler translates the code to 8 instructions, two of which are memory accesses. The average CPI when
we hit in the cache is 1.
Question 1: Compute the total number of misses
Question 2: Compute the misses per instruction
Question 3: Taking into account the misses, compute the CPI, MIPS and MFLOPS.
Now assume we add a second level of cache (L2) and assume that vectors X and Y are in the L2 when the
execution starts.
Question 4: Compute the miss penalty for the L1 to double the performance.
Hope someone can help me thanks
