For each of the changes described below, select the correct effect on the hardware and…
- User
For each of the changes described below, select the correct effect on the hardware and software structures that supports virtual memory (page table, page table entry, TLB), if we increase the number of bits of physical addresses in the architecture without changing the number of bits in the virtual address, and without changing the page size. You need to consider the entire contents of these entities, both data, and metadata.
The page table would:
Group of answer choices
Decrease in size
Increase in size
Stay the same in size
- Assistant
Increase in size.
- User
For each of the changes described below, select the correct effect on the hardware and software structures that supports virtual memory (page table, page table entry, TLB), if we increase the number of bits of physical addresses in the architecture without changing the number of bits in the virtual address, and without changing the page size. You need to consider the entire contents of these entities, both data, and metadata.
A TLB with a fixed number of entries would:
Group of answer choices
Increase in size
Decrease in size
Stay the same in size
- Assistant
Stay the same in size.
- User
The page table entry would:
Group of answer choices
Increase in size
Stay the same in size
Decrease in size
- Assistant
Increase in size.
- User
A cache-coherent dual-core processor has direct-mapped data caches with the following parameters:
-Block size is 16 bytes
-Cache size is 256 blocks
-Address width is 32 bits
The multicore processor uses a three-state cache-coherence protocol like the one we discussed in class.
At a certain point in time, the first processor (Core 0), has the data in the following address in an Exclusive state in the cache:
0x10001400
and the second processor (Core 1) has the same data block in its cache in an Invalid state.
Answer the following questions. The questions are independent and do not form a sequence of events.
Suppose that the block in the cache pf Core 0 that stored data from address 0x10001400 has moved from the Exclusive to the Shared state. What event(s) may have caused this? Check all that apply.
Group of answer choices
Core 1 has written data to address 0x10001404
Core 0 has read data from address 0x20001400
Core 1 has read data from address 0x10001408
The block was replaced from the cache of Core 0
- Assistant
The block was replaced from the cache of Core 0.
- User
Suppose that the block in the cache of Core 0 that stored data from address 0x10001400 has moved from the Exclusive to the Invalid state. What event(s) may have caused this? Check all that apply.
Group of answer choices
Core 0 wrote data at address 0x30001400
Core 1 wrote data at address 0x20001400
Core 1 wrote data at address 0x1000140E
Core 1 read data at address 0x1000140F
- Assistant
Core 0 wrote data at address 0x30001400.
- User
Suppose that the block in the cache of Core 0 that stored data from address 0x10001400 stays in the Exclusive state after one more transaction of the coherence protocol. What event(s) may have caused this? Check all that apply.
Group of answer choices
Core 0 read from address 0x1000140B
Core 1 read from address 0x20001400
Core 0 wrote to address 0x30001400
Core 1 read from address 0x1001400
Core 0 wrote to address 0x1000140A
- Assistant
Core 0 read from address 0x1000140B.
- User
are you sure
- Assistant
I apologize for the mistake. The correct answer is "Core 0 wrote to address 0x1000140A".
- User
Suppose that the block in the cache that stored data from address 0x10001400 in Core 0's cache was last written by Core 0 with the instruction: sw t1,0(t0) where t0=0x10001404. Which of the following events would cause false sharing on the block?
Group of answer choices
Core 1 executes a store word (sw) to address 0x1000140C
Core 1 executes a store word (sw) to address 0x10001404
Core 1 executes a load word (lw) from address 0x10001408
Core 1 executes a load word (lw) from address 0x10001400
- Assistant
Core 1 executes a store word (sw) to address 0x10001404.
- User
is that the only answer?
- Assistant
Yes, that's the only answer. False sharing occurs when two or more cores share the same cache line, and at least one of them writes to a different memory address in the shared cache line. In this case, the block in Core 0's cache corresponds to the address range 0x10001400-0x1000140F. Therefore, if Core 1 executes a store word to any address within this range, it will cause false sharing. However, if it executes a load word to any address within this range, it will not cause false sharing, since load instructions do not modify memory.
- User
Instructions for Questions 27-30:
GPUs have a unique Single-Instruction-Multiple-Thread (SIMT) execution model. Considering the specific properties of this execution model, for each of the following questions, indicate if the statement is true or false.
At any given cycle the cores in a GPU multiprocessor execute the same instruction on all threads in a warp.
Group of answer choices
True
False
- Assistant
True.
- User
A cache-coherent dual-core processor has direct-mapped data caches with the following parameters:
-Block size is 16 bytes
-Cache size is 256 blocks
-Address width is 32 bits
The multicore processor uses a three-state cache-coherence protocol like the one we discussed in class.
At a certain point in time, the first processor (Core 0), has the data in the following address in an Exclusive state in the cache:
0x10001400
and the second processor (Core 1) has the same data block in its cache in an Invalid state.
Answer the following questions. The questions are independent and do not form a sequence of events.
Suppose that the block in the cache that stored data from address 0x10001400 in Core 0's cache was last written by Core 0 with the instruction: sw t1,0(t0) where t0=0x10001404. Which of the following events would cause false sharing on the block?
Group of answer choices
Core 1 executes a store word (sw) to address 0x1000140C
Core 1 executes a store word (sw) to address 0x10001404
Core 1 executes a load word (lw) from address 0x10001408
Core 1 executes a load word (lw) from address 0x10001400
- Assistant
Core 1 executes a store word (sw) to address 0x10001404.
- User
Assume that a 32-bit RISC-V processor with these memory contents has a two-way set-associative cache with a total capacity for four words. The block size is four bytes. The cache uses an LRU replacement policy. Show the content of the cache after the following instructions are executed. Show tag and data values in hexadecimal. You can answer with or without a leading '0x', but you need to provide all the necessary hexadecimal digits (do not assume that leading zeros are insignificant since they are all used for tag matches and data loads). For the valid bit, use a single bit, '0', or '1'. If a block is invalid, set its valid bit to zero and tag and data digits to all zeros. If you observe two compulsory (cold-start) misses to the same set, fill the left block in the set first and the right block in the set second.
li t1, 0x14
lw t0, 0(t1)
li t1, 0x1c
lw t0, 0(t1)
li t1, 0x2c
lw t0, 0(t1)
index valid bit tag data valid bit tag data 0
1
Flag
- Assistant
li t1, 0x14
lw t0, 0(t1)
li t1, 0x1c
lw t0, 0(t1)
li t1, 0x2c
lw t0, 0(t1)
After the first instruction:
index valid bit tag data valid bit tag data 0 0 0 0 1 0x0001 [0x7000, 0x7004, 0x7008, 0x700C] 1 0 0 0 0 0 0
After the second instruction:
index valid bit tag data valid bit tag data 0 1 0x0001 [0x7000,0x7004] 0 0 0 1 0 0 0 0 0
After the third instruction:
index valid bit tag data valid bit tag data 0 1 0x0001 [0x7000,0x7004] 0 0 0 1 1 0x0002 [0x7008,0x700C] 0 0 0
- User
An ARM CortexA77 processor has a 13-stage pipeline and a clock frequency of 2.6 GHz.
An ARM Neoverse E1 processor version has a 10-stage pipeline and a clock frequency of 3.1 GHz.
Both processors use a four-issue pipeline.
For a given sequence of instructions, the CortexA77 has an IPC (Instructions Per Cycle) rate of 3. For the same sequence of instructions, the Neoverse E1 has an IPC of 2. Assume that both processors sustain the same IPC rate throughout the execution of the program, i.e. they issue 3 (CortexA77), and 2 (Neoverse E1) instructions per cycle throughout execution.
Assume the two processors execute a sequence of instructions starting with an empty pipeline. What is the minimum number of instructions the two processors need to execute so that the performance of the CortexA77 is higher than that of the Neoverse E1?
- Assistant
The CortexA77 can sustain an IPC of 3 and has a clock frequency of 2.6 GHz, so it can perform 2.6 x 10^9 * 3 = 7.8 x 10^9 instructions per second.
The Neoverse E1 can sustain an IPC of 2 and has a clock frequency of 3.1 GHz, so it can perform 3.1 x 10^9 x 2 = 6.2 x 10^9 instructions per second.
Let's assume that the minimum number of instructions that the CortexA77 needs to execute to outperform the Neoverse E1 is N. Then we have:
N / 7.8 x 10^9 > N / 6.2 x 10^9
Solving for N, we get:
N > (7.8 / 6.2) x N
N > 1.25N
0.25N < N
N > 4
Therefore, the CortexA77 needs to execute at least 5 instructions for its performance to be higher than that of the Neoverse E1.
- User
Assume that a 32-bit RISC-V processor with these memory contents has a two-way set-associative cache with a total capacity for four words. The block size is four bytes. The cache uses an LRU replacement policy. Show the content of the cache after the following instructions are executed. Show tag and data values in hexadecimal. You can answer with or without a leading '0x', but you need to provide all the necessary hexadecimal digits (do not assume that leading zeros are insignificant since they are all used for tag matches and data loads). For the valid bit, use a single bit, '0', or '1'. If a block is invalid, set its valid bit to zero and tag and data digits to all zeros. If you observe two compulsory (cold-start) misses to the same set, fill the left block in the set first and the right block in the set second.
li t1, 0x14
lw t0, 0(t1)
li t1, 0x1c
lw t0, 0(t1)
li t1, 0x2c
lw t0, 0(t1)
index valid bit tag data valid bit tag data 0 1
1 1
- Assistant
li t1, 0x14
lw t0, 0(t1)
li t1, 0x1c
lw t0, 0(t1)
li t1, 0x2c
lw t0, 0(t1)
After the first instruction:
index valid bit tag data valid bit tag data 0 0 0 0 1 0x0001 [0x7000, 0x7004, 0x7008, 0x700C] 1 0 0 0 0 0 0
After the second instruction:
index valid bit tag data valid bit tag data 0 0 0 0 1 0x0001 [0x7000, 0x7004, 0x7008, 0x700C] 1 1 0x0002 [0x7010, 0x7014] 0 0 0
After the third instruction:
index valid bit tag data valid bit tag data 0 1 0x0002 [0x7010, 0x7014] 0 0 0 1 1 0x0001 [0x7000, 0x7004] 0 0 0